A method, device, and medium for determining thermal simulation parameters

Abstract

The embodiment of the application discloses a kind of determination methods, device, equipment and medium of thermal simulation parameter, related to data processing technical field.The determination method of thermal simulation parameter provided in the embodiment of the application can determine at least one of power density distribution, core particle area, convective heat transfer coefficient and thermal conductivity adapted to the target temperature distribution by inputting the target temperature distribution required into the second heat flow simulation neural network model trained, so that the temperature distribution of system-level chip can be regulated according to at least one of power density distribution, core particle area, convective heat transfer coefficient and thermal conductivity in actual application, and then the heat dissipation design is optimized, and the reliability and performance of system-level chip are improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, equipment and medium for determining thermal simulation parameters. Background Technology

[0002] As the semiconductor industry enters the "post-Moore's Law era," traditional methods of improving chip performance by shrinking transistor size are approaching their physical limits. To continue the trend of chip performance improvement, three-dimensional heterogeneous integration technology has become an important development direction for the semiconductor industry. Three-dimensional heterogeneous integration technology refers to integrating chips (or dies) with different processes and functions into a single package through three-dimensional stacking or high-density interconnection to form a high-performance, multifunctional, low-power system-on-a-chip (SoC).

[0003] However, this high-density integration also brings severe heat dissipation challenges: heat is easily accumulated inside the system-on-a-chip, forming local "hot spots," which leads to a decrease in heat dissipation efficiency and thus affects the stability and performance of the system-on-a-chip.

[0004] Therefore, how to optimize heat dissipation design by regulating the temperature distribution of system-on-a-chip (SoC) and improve the reliability and performance of SoC has become an urgent technical problem to be solved. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a method, apparatus, device, and medium for determining thermal simulation parameters, which can optimize heat dissipation design and improve the reliability and performance of system-on-a-chip by adjusting the temperature distribution of the system-on-a-chip.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] In a first aspect, this application discloses a method for determining thermal simulation parameters, the method comprising:

[0008] Build a heterogeneous integration architecture;

[0009] Based on the heterogeneous integrated architecture, a first training dataset is generated, wherein the first training dataset includes thermal simulation parameters of the heterogeneous integrated architecture, and simulated temperature distribution obtained by running thermal simulation on the heterogeneous integrated architecture according to the thermal simulation parameters. The thermal simulation parameters include at least one of architecture size, heat source parameters, material properties, boundary conditions and initial conditions.

[0010] Using the thermal simulation parameters as input and the simulated temperature distribution as output, a first thermal flux simulation neural network model is trained.

[0011] Based on the first heat flux simulation neural network model, a second training dataset is generated, wherein the second training dataset includes the simulated temperature distribution, and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution;

[0012] Using the simulated temperature distribution as input, and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution as output, a second heat flow simulation neural network model is trained to obtain the second heat flow simulation neural network model.

[0013] By inputting the target temperature distribution into the second heat flux simulation neural network model, at least one of the following is determined: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity corresponding to the target temperature distribution.

[0014] Optionally, the step of training a first heat flux simulation neural network model by taking the thermal simulation parameters as input and the simulated temperature distribution as output includes:

[0015] The thermal simulation parameters are input into the neural network model to obtain the predicted temperature distribution output by the neural network model.

[0016] The model parameters of the neural network model are updated by calculating the loss function between the predicted temperature distribution and the simulated temperature distribution;

[0017] Repeat the step of inputting the thermal simulation parameters into the neural network model until the loss function converges to obtain the first thermal flux simulation neural network model.

[0018] Optionally, the neural network model has a multilayer perceptron architecture; the relationship between the output and input of the first heat flux simulation neural network model can be expressed by the following formula:

[0019]

[0020] Where T represents the predicted temperature distribution. This represents the composite operation of a neural network, where σ is the activation function and N is the number of nodes. k (X (k) ) indicates that the operation of the k-th layer of the neural network is performed.

[0021] Optionally, the simulated temperature distribution in the second training dataset is less than a preset temperature distribution threshold.

[0022] Optionally, the construction of the heterogeneous integration architecture includes:

[0023] Constructing a primitive architecture, wherein the primitive architecture includes at least one of a chip, a microbump, an underfill material, a redistribution layer, a through-silicon via, and a wafer substrate;

[0024] By converting the anisotropic materials in the original architecture into isotropic materials, a heterogeneous integrated architecture is obtained.

[0025] Secondly, this application discloses a device for determining thermal simulation parameters, the device comprising: an architecture building module, a first generation module, a first training module, a second generation module, a second training module, and a parameter determination module;

[0026] The architecture building module is used to build a heterogeneous integrated architecture;

[0027] The first generation module is used to generate a first training dataset based on the heterogeneous integrated architecture, wherein the first training dataset includes thermal simulation parameters of the heterogeneous integrated architecture and simulated temperature distribution obtained by running thermal simulation on the heterogeneous integrated architecture according to the thermal simulation parameters, and the thermal simulation parameters include at least one of architecture size, heat source parameters, material properties, boundary conditions and initial conditions.

[0028] The first training module is used to take the thermal simulation parameters as input and the simulated temperature distribution as output to train a first thermal flux simulation neural network model.

[0029] The second generation module is used to generate a second training dataset based on the first heat flux simulation neural network model, wherein the second training dataset includes the simulated temperature distribution, and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution;

[0030] The second training module is used to take the simulated temperature distribution as input and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution as output, and train a second heat flow simulation neural network model.

[0031] The parameter determination module is used to determine at least one of the following: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity, corresponding to the target temperature distribution, by inputting the target temperature distribution into the second heat flux simulation neural network model.

[0032] Optionally, the first training module includes: a first training submodule, a second training submodule, and a third training submodule;

[0033] The first training submodule is used to input the thermal simulation parameters into the neural network model to obtain the predicted temperature distribution output by the neural network model;

[0034] The second training submodule is used to update the model parameters of the neural network model by calculating the loss function between the predicted temperature distribution and the simulated temperature distribution;

[0035] The third training submodule is used to repeatedly execute the step of inputting the thermal simulation parameters into the neural network model until the loss function converges, thereby obtaining the first thermal flow simulation neural network model.

[0036] Optionally, the neural network model has a multilayer perceptron architecture; the relationship between the output and input of the first heat flux simulation neural network model can be expressed by the following formula:

[0037]

[0038] Where T represents the predicted temperature distribution. This represents the composite operation of a neural network, where σ is the activation function and N is the number of nodes. k (X (k) ) indicates that the operation of the k-th layer of the neural network is performed.

[0039] Optionally, the simulated temperature distribution in the second training dataset is less than a preset temperature distribution threshold.

[0040] Optionally, the architecture building module includes: a first building submodule and a second building submodule;

[0041] The first construction submodule is used to construct the original architecture, wherein the original architecture includes at least one of the following: a chip, a microbump, an underfill material, a redistribution layer, a through-silicon via, and a wafer substrate;

[0042] The second construction submodule is used to obtain a heterogeneous integrated architecture by converting the anisotropic materials in the original architecture into isotropic materials.

[0043] Thirdly, this application discloses a device for determining thermal simulation parameters, the device comprising: a memory and a processor;

[0044] The memory is used to store programs;

[0045] The processor is configured to execute the program to implement the steps of the method for determining thermal simulation parameters as described in the first aspect.

[0046] Fourthly, this application discloses a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for determining thermal simulation parameters as described in the first aspect.

[0047] Compared with the prior art, this application has the following beneficial effects:

[0048] This application discloses a method, apparatus, device, and medium for determining thermal simulation parameters. The method includes: constructing a heterogeneous integrated architecture; generating a first training dataset based on the heterogeneous integrated architecture, wherein the first training dataset includes thermal simulation parameters of the heterogeneous integrated architecture and a simulated temperature distribution obtained by running thermal simulation on the heterogeneous integrated architecture according to the thermal simulation parameters, wherein the thermal simulation parameters include at least one of architecture size, heat source parameters, material properties, boundary conditions, and initial conditions; training a first thermal flux simulation neural network model by using the thermal simulation parameters as input and the simulated temperature distribution as output; generating a second training dataset based on the first thermal flux simulation neural network model, wherein the second training dataset includes a simulated temperature distribution and at least one of power density distribution, core area, convective heat transfer coefficient, and thermal conductivity corresponding to the simulated temperature distribution; training a second thermal flux simulation neural network model by using the simulated temperature distribution as input and at least one of power density distribution, core area, convective heat transfer coefficient, and thermal conductivity corresponding to the simulated temperature distribution as output; and determining at least one of power density distribution, core area, convective heat transfer coefficient, and thermal conductivity corresponding to the target temperature distribution by inputting a target temperature distribution into the second thermal flux simulation neural network model. Therefore, the method for determining thermal simulation parameters provided in this application can determine at least one of the following: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity, which are adapted to the target temperature distribution, by inputting the required target temperature distribution into the trained second thermal flux simulation neural network model. In practical applications, the temperature distribution of the system-on-a-chip can be adjusted according to at least one of the following: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity, thereby optimizing the heat dissipation design and improving the reliability and performance of the system-on-a-chip. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0050] Figure 1 A flowchart illustrating a method for determining thermal simulation parameters provided in this application embodiment;

[0051] Figure 2 A schematic diagram of a heterogeneous integrated architecture provided in an embodiment of this application;

[0052] Figure 3 A schematic diagram of a first heat flux simulation neural network model provided in an embodiment of this application;

[0053] Figure 4 A schematic diagram of a second heat flux simulation neural network model provided in an embodiment of this application;

[0054] Figure 5 A schematic diagram of a device for determining thermal simulation parameters provided in an embodiment of this application;

[0055] Figure 6 This is a schematic diagram of a computer-readable medium provided in an embodiment of this application. Detailed Implementation

[0056] As described earlier, to continue the trend of improving chip performance, three-dimensional heterogeneous integration technology has become an important development direction in the semiconductor industry. However, high-density integration also brings severe heat dissipation challenges: heat tends to accumulate inside the system-on-a-chip (SoC), forming localized "hot spots," leading to decreased heat dissipation efficiency and consequently affecting the stability and performance of the SoC. Therefore, how to optimize the heat dissipation design by controlling the temperature distribution of the SoC and improving its reliability and performance has become an urgent technical problem to be solved.

[0057] Through research, the inventors have proposed a method, apparatus, device, and medium for determining thermal simulation parameters. The method for determining thermal simulation parameters provided in this application, by inputting the desired target temperature distribution into a trained second thermal flux simulation neural network model, can determine at least one of the following: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity, which is adapted to the target temperature distribution. In practical applications, the temperature distribution of the system-on-a-chip can be adjusted according to at least one of the following: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity, thereby optimizing the heat dissipation design and improving the reliability and performance of the system-on-a-chip.

[0058] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0059] See Figure 1 The figure is a flowchart of a method for determining thermal simulation parameters provided in an embodiment of this application. The method includes:

[0060] S101: Build a heterogeneous integrated architecture.

[0061] In some specific implementation methods, the construction of heterogeneous integration architecture is carried out as follows:

[0062] Construct a primitive architecture, wherein the primitive architecture includes at least one of a chiplet, a microbump, an underfill material, a redistribution layer (including RDL1 and RDL2), a through-silicon via (TSV), and a wafer substrate.

[0063] Because the original architecture is quite complex (e.g., numerous microbumps, tiny TSV size) and each component exhibits significant anisotropic thermal properties, directly modeling the original architecture would require an extremely large amount of computation, resulting in very low simulation efficiency. Therefore, this application addresses this by both equipping the fine interconnect architecture in the original architecture with equivalents and equipping the anisotropic materials in the original architecture with isotropic materials, thereby obtaining a heterogeneous integrated architecture. See [link to relevant documentation] Figure 2 The figure is a schematic diagram of a heterogeneous integrated architecture provided in an embodiment of this application. This not only reduces the complexity of the heterogeneous integrated architecture, but also improves the simulation efficiency of the heterogeneous integrated architecture.

[0064] S102: Based on the heterogeneous integrated architecture, generate a first training dataset, wherein the first training dataset includes thermal simulation parameters of the heterogeneous integrated architecture, and simulated temperature distribution obtained by running thermal simulation on the heterogeneous integrated architecture according to the thermal simulation parameters. The thermal simulation parameters include at least one of architecture size, heat source parameters, material properties, boundary conditions and initial conditions.

[0065] In one specific implementation, based on a heterogeneous integration architecture, the method for generating the first training dataset is as follows:

[0066] First, thermal simulation parameters for the heterogeneous integrated architecture are defined using thermal simulation tools (such as ANSYS, COMSOL, etc.). These thermal simulation parameters include at least one of the following: architecture dimensions, heat source parameters, material properties, boundary conditions, and initial conditions. Specifically, architecture dimensions include chip area (A), TSV diameter, and redistribution layer thickness; heat source parameters include power density distribution (P0) and peak power consumption; material properties include thermal conductivity (k), specific heat capacity (c), and density (ρ); boundary conditions include cooling method (convective heat transfer coefficient h) and ambient temperature (T∞); and initial conditions include the initial temperature field (T0) during chip startup.

[0067] Secondly, based on the defined heterogeneous integrated architecture, the temperature prediction function of the thermal simulation tool is invoked to solve the heat conduction equation (i.e., run the thermal simulation) to generate a simulated temperature distribution. For example, the simulated temperature distribution can be a temperature distribution cloud map T(x,y,z) of the entire chip.

[0068] Subsequently, by changing the above thermal simulation parameters, the simulated temperature distribution under different thermal simulation parameters was obtained, and a set of "thermal simulation parameters - simulated temperature distribution" was constructed.

[0069] Finally, a first training dataset is selected from the set of "thermal simulation parameters - simulated temperature distribution". This first training dataset is used to train the neural network model. For example, the first training dataset can comprise 70%-80% of the set of "thermal simulation parameters - simulated temperature distribution". It should be noted that this application does not limit the specific percentage.

[0070] It should also be noted that the remaining 20%-30% of the "thermal simulation parameters - simulated temperature distribution" set will be used as the first test dataset to verify the generalization ability of the neural network model and avoid overfitting.

[0071] S103: Using thermal simulation parameters as input and simulated temperature distribution as output, the first thermal flux simulation neural network model is trained.

[0072] See Figure 3 This figure is a schematic diagram of a first heat flux simulation neural network model provided in an embodiment of this application. Figure 3 As can be seen, the neural network architecture of the first thermal flux simulation neural network model is a multilayer perceptron (MLP). The MLP includes an input layer (for receiving thermal simulation parameters), hidden layers (containing M layers, each with n neurons, using an activation function σ to introduce nonlinearity), and an output layer (for outputting the simulated temperature distribution T). It is understandable that the activation function of the output layer can generally be omitted, and a linear output is used. For example, the input of the first thermal flux simulation neural network model can be shown in Table 1 below, and the output of the first thermal flux simulation neural network model is the simulated temperature distribution T.

[0073] Table 1

[0074]

[0075] In one specific implementation, the thermal simulation parameters are used as input, and the simulated temperature distribution is used as output to train the first thermal flux simulation neural network model. The specific implementation method is as follows:

[0076] First, the thermal simulation parameters are input into the neural network model to obtain the predicted temperature distribution T' output by the neural network model.

[0077] Subsequently, the model parameters of the neural network model are updated by calculating the loss function between the predicted temperature distribution T' and the simulated temperature distribution T corresponding to the thermal simulation parameters in the first training dataset. For example, the loss function can be the mean squared error loss function, but this application does not limit it to this.

[0078] Finally, the steps of inputting thermal simulation parameters into the neural network model are repeated until the loss function converges, thus obtaining the first thermal flux simulation neural network model. At this point, the relationship between the output and input of the first thermal flux simulation neural network model can be expressed by the following formula (1):

[0079]

[0080] Where T represents the predicted temperature distribution. This represents the operation of a neural network, where σ is the activation function and N is the number of nodes. k (X (k) ) indicates performing the operation of the k-th layer of the neural network, N k (X (k) ) = W k X+b k W k Let b be the weight matrix. k This is the bias vector.

[0081] It should be noted that during training, the performance of the first heat flux simulation neural network model can be evaluated periodically (e.g., every 100 iterations) using the first validation set data to monitor for overfitting. If overfitting occurs, the current first heat flux simulation neural network model should be saved as the optimal model immediately, i.e., early stopping should be initiated.

[0082] S104: Based on the first heat flux simulation neural network model, generate a second training dataset, wherein the second training dataset includes a simulated temperature distribution, and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution.

[0083] From the simulated temperature distribution output by the first thermal flux simulation neural network model, a simulated temperature distribution below a preset temperature distribution threshold (e.g., 85°C) is selected. This is because if the temperature of the system-on-a-chip exceeds the preset temperature distribution threshold, it is highly likely to cause reliability problems (such as thermal failure or performance degradation). Therefore, a second training dataset is generated based on the simulated temperature distribution below the preset temperature threshold, and at least one of the power density distribution, chip area, convective heat transfer coefficient, and thermal conductivity corresponding to the simulated temperature distribution.

[0084] It should be noted that during the generation of the second training dataset, oversampling can be performed near a preset temperature distribution threshold (e.g., 83-85℃) to enhance the predictive ability of the subsequent second heat flux simulation neural network model for critical states. This application does not impose any limitations on this aspect.

[0085] S105: Based on the model architecture of the first heat flux simulation neural network model, at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution is used as the output to train the second heat flux simulation neural network model.

[0086] The method for determining thermal simulation parameters provided in this application utilizes a neural network inverse mapping (the input is the simulated temperature distribution, and the output is at least one of power density distribution, core area, convective heat transfer coefficient, and thermal conductivity) to achieve automated and intelligent control. See also Figure 4 The figure is a schematic diagram of a second heat flux simulation neural network model provided in an embodiment of this application. The input and output of the second heat flux simulation neural network model are shown in Table 2 below:

[0087] Table 2

[0088]

[0089] Understandably, the second thermal flux simulation neural network adopts the multilayer perceptron (MLP) architecture design of the first thermal flux simulation neural network. The two maintain similarities in model structure, but may differ in hyperparameter configurations such as model width.

[0090] S106: By inputting the target temperature distribution into the second heat flux simulation neural network model, determine at least one of the power density distribution, core area, convective heat transfer coefficient, and thermal conductivity corresponding to the target temperature distribution.

[0091] In practical applications, by inputting the desired target temperature distribution (e.g., 60°C) into the trained second heat flux simulation neural network model, at least one of the power density distribution, core area, convective heat transfer coefficient, and thermal conductivity output by the second heat flux simulation neural network model corresponding to the target temperature distribution can be determined.

[0092] In summary, this application discloses a method for determining thermal simulation parameters. The method provided in this application determines at least one of the following thermal simulation parameters by inputting the desired target temperature distribution into a trained second thermal flux simulation neural network model: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity. In practical applications, the temperature distribution of the system-on-a-chip can be adjusted according to at least one of the following parameters: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity, thereby optimizing the heat dissipation design and improving the reliability and performance of the system-on-a-chip.

[0093] See Figure 5The figure is a schematic diagram of a thermal simulation parameter determination device provided in an embodiment of this application. The thermal simulation parameter determination device 500 includes: an architecture construction module 501, a first generation module 502, a first training module 503, a second generation module 504, a second training module 505, and a parameter determination module 506;

[0094] Architecture building module 501 is used to build heterogeneous integrated architectures;

[0095] The first generation module 502 is used to generate a first training dataset based on the heterogeneous integrated architecture. The first training dataset includes thermal simulation parameters of the heterogeneous integrated architecture and simulated temperature distribution obtained by running thermal simulation on the heterogeneous integrated architecture according to the thermal simulation parameters. The thermal simulation parameters include at least one of architecture size, heat source parameters, material properties, boundary conditions and initial conditions.

[0096] The first training module 503 is used to train the first heat flow simulation neural network model by taking thermal simulation parameters as input and simulated temperature distribution as output.

[0097] The second generation module 504 is used to generate a second training dataset based on the first heat flux simulation neural network model. The second training dataset includes a simulated temperature distribution and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution.

[0098] The second training module 505 is used to train the second heat flow simulation neural network model based on the model architecture of the first heat flow simulation neural network model, taking the simulated temperature distribution as input and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution as output.

[0099] The parameter determination module 506 is used to determine at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the target temperature distribution by inputting the target temperature distribution into the second heat flux simulation neural network model.

[0100] In one specific implementation, the first training module 503 includes: a first training submodule, a second training submodule, and a third training submodule;

[0101] The first training submodule is used to input thermal simulation parameters into the neural network model and obtain the predicted temperature distribution output by the neural network model.

[0102] The second training submodule is used to update the model parameters of the neural network model by calculating the loss function between the predicted temperature distribution and the simulated temperature distribution.

[0103] The third training submodule is used to repeatedly execute the step of inputting thermal simulation parameters into the neural network model until the loss function converges, thus obtaining the first thermal flow simulation neural network model.

[0104] In one specific implementation, the neural network model is a multilayer perceptron architecture; the relationship between the output and input of the first heat flux simulation neural network model can be shown by the following formula (2):

[0105]

[0106] Where T represents the predicted temperature distribution. This represents the composite operation of a neural network, where σ is the activation function and N is the number of nodes. k (X (k) ) indicates that the operation of the k-th layer of the neural network is performed.

[0107] In one specific implementation, the simulated temperature distribution in the second training dataset is less than a preset temperature distribution threshold.

[0108] In one specific implementation, the architecture building module 501 includes: a first building submodule and a second building submodule;

[0109] The first construction submodule is used to construct the original architecture, wherein the original architecture includes at least one of the following: a chip, a microbump, an underfill material, a redistribution layer, a through-silicon via, and a wafer substrate;

[0110] The second construction submodule is used to obtain a heterogeneous integrated architecture by converting the anisotropic materials in the original architecture into isotropic materials.

[0111] In summary, this application discloses a device for determining thermal simulation parameters. The device for determining thermal simulation parameters provided in the embodiments of this application can determine at least one of the following: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity, by inputting the required target temperature distribution into a trained second thermal flux simulation neural network model. In practical applications, the temperature distribution of the system-on-a-chip can be adjusted according to at least one of the following: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity, thereby optimizing the heat dissipation design and improving the reliability and performance of the system-on-a-chip.

[0112] This application also provides a device for determining corresponding thermal simulation parameters and a computer-readable medium for implementing the method for determining thermal simulation parameters provided in this application.

[0113] The device for determining thermal simulation parameters includes a memory and a processor. The memory is used to store instructions or code, and the processor is used to execute the instructions or code so that the device performs a method for determining thermal simulation parameters according to any embodiment of this application.

[0114] See Figure 6 This figure is a schematic diagram of a computer-readable medium provided in an embodiment of this application. The computer-readable medium 600 stores a computer program 611, which, when executed by a processor, implements the above-described... Figure 1 The steps for determining the thermal simulation parameters.

[0115] It should be noted that, in the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0116] It should be noted that the machine-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0117] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0118] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

[0119] While several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this application. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0120] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for determining thermal simulation parameters, characterized in that, The method includes: Build a heterogeneous integration architecture; Based on the heterogeneous integrated architecture, a first training dataset is generated, wherein the first training dataset includes thermal simulation parameters of the heterogeneous integrated architecture, and simulated temperature distribution obtained by running thermal simulation on the heterogeneous integrated architecture according to the thermal simulation parameters. The thermal simulation parameters include at least one of architecture size, heat source parameters, material properties, boundary conditions and initial conditions. Using the thermal simulation parameters as input and the simulated temperature distribution as output, a first thermal flux simulation neural network model is trained. Based on the first heat flux simulation neural network model, a second training dataset is generated, wherein the second training dataset includes the simulated temperature distribution, and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution; Using the simulated temperature distribution as input, and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution as output, a second heat flow simulation neural network model is trained to obtain the second heat flow simulation neural network model. By inputting the target temperature distribution into the second heat flux simulation neural network model, at least one of the following is determined: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity corresponding to the target temperature distribution.

2. The method according to claim 1, characterized in that, The step of training a first heat flow simulation neural network model by taking the thermal simulation parameters as input and the simulated temperature distribution as output includes: The thermal simulation parameters are input into the neural network model to obtain the predicted temperature distribution output by the neural network model. The model parameters of the neural network model are updated by calculating the loss function between the predicted temperature distribution and the simulated temperature distribution; Repeat the step of inputting the thermal simulation parameters into the neural network model until the loss function converges to obtain the first thermal flux simulation neural network model.

3. The method according to claim 2, characterized in that, The neural network model has a multilayer perceptron architecture; the relationship between the output and input of the first heat flux simulation neural network model can be expressed by the following formula: Where T represents the predicted temperature distribution. This represents the composite operation of a neural network, where σ is the activation function and N is the number of nodes. k (X (k) ) indicates that the operation of the k-th layer of the neural network is performed.

4. The method according to claim 1, characterized in that, The simulated temperature distribution in the second training dataset is less than a preset temperature distribution threshold.

5. The method according to claim 1, characterized in that, The construction of the heterogeneous integration architecture includes: Constructing a primitive architecture, wherein the primitive architecture includes at least one of a chip, a microbump, an underfill material, a redistribution layer, a through-silicon via, and a wafer substrate; By converting the anisotropic materials in the original architecture into isotropic materials, a heterogeneous integrated architecture is obtained.

6. A device for determining thermal simulation parameters, characterized in that, The device includes: an architecture construction module, a first generation module, a first training module, a second generation module, a second training module, and a parameter determination module; The architecture building module is used to build a heterogeneous integrated architecture; The first generation module is used to generate a first training dataset based on the heterogeneous integrated architecture, wherein the first training dataset includes thermal simulation parameters of the heterogeneous integrated architecture and simulated temperature distribution obtained by running thermal simulation on the heterogeneous integrated architecture according to the thermal simulation parameters, and the thermal simulation parameters include at least one of architecture size, heat source parameters, material properties, boundary conditions and initial conditions. The first training module is used to take the thermal simulation parameters as input and the simulated temperature distribution as output to train a first thermal flux simulation neural network model. The second generation module is used to generate a second training dataset based on the first heat flux simulation neural network model, wherein the second training dataset includes the simulated temperature distribution, and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution; The second training module is used to take the simulated temperature distribution as input and at least one of the power density distribution, core area, convective heat transfer coefficient and thermal conductivity corresponding to the simulated temperature distribution as output, and train a second heat flow simulation neural network model. The parameter determination module is used to determine at least one of the following: power density distribution, core area, convective heat transfer coefficient, and thermal conductivity, corresponding to the target temperature distribution, by inputting the target temperature distribution into the second heat flux simulation neural network model.

7. The apparatus according to claim 6, characterized in that, The first training module includes: a first training submodule, a second training submodule, and a third training submodule; The first training submodule is used to input the thermal simulation parameters into the neural network model to obtain the predicted temperature distribution output by the neural network model; The second training submodule is used to update the model parameters of the neural network model by calculating the loss function between the predicted temperature distribution and the simulated temperature distribution; The third training submodule is used to repeatedly execute the step of inputting the thermal simulation parameters into the neural network model until the loss function converges, thereby obtaining the first thermal flow simulation neural network model.

8. The apparatus according to claim 7, characterized in that, The neural network model has a multilayer perceptron architecture; the relationship between the output and input of the first heat flux simulation neural network model can be expressed by the following formula: Where T represents the predicted temperature distribution. This represents the composite operation of a neural network, where σ is the activation function and N is the number of nodes. k (X (k) ) indicates that the operation of the k-th layer of the neural network is performed.

9. A device for determining thermal simulation parameters, characterized in that, The device includes: a memory and a processor; The memory is used to store programs; The processor is configured to execute the program to implement each step of the method for determining thermal simulation parameters as described in any one of claims 1 to 5.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements each step of the method for determining thermal simulation parameters as described in any one of claims 1 to 5.

Images