Core particle layout inversion model training method and core particle layout determination method

Abstract

The invention discloses a core particle layout inversion model training method and a core particle layout determination method. The method comprises the following steps: acquiring temperature distribution data of a target heterogeneous integrated system; the temperature distribution data represents the temperature distribution condition of the heterogeneous integrated system at the same moment; performing inversion analysis on the temperature distribution data by adopting a trained core particle layout inversion model, and determining target core particle layout data of the target heterogeneous integrated system, so that a user performs layout optimization processing on the target heterogeneous integrated system according to the target core particle layout data; the trained core particle layout inversion model is obtained by performing first model training on a preset core particle layout inversion model by using the simulation data set and then performing second model training by using the experimental data set. According to the method provided by the invention, the core particle layout inversion model is trained by using the simulation data and the experimental data so as to accurately invert the core particle layout data based on the temperature distribution data, and the pertinence and the thermal management effect of core particle layout optimization are improved.

Description

Technical Field

[0001] The present application relates to the field of integrated circuit technology, and in particular, to a method for training a die layout inversion model and a method for determining a die layout. Background Art

[0002] With the rapid development of electronic technology, high-density integration technology represented by heterogeneous integration systems has become a key means to improve the performance of electronic devices. However, due to the different power consumption and heat generation characteristics of different dies, the temperature distribution inside the system is uneven, and the local hot spot phenomenon is serious, so it is necessary to optimize the die layout of the heterogeneous integration system.

[0003] In a heterogeneous integration system, since the heat source positions and power distributions of different dies are often complex and variable, traditional methods often rely on basic information such as the power consumption and operating frequency of the die in the die design document and specification book, and then estimate the approximate heat source position and power distribution of the die and other die layout information, which leads to a lack of pertinence in optimizing the die layout and poor thermal management effect. Summary of the Invention

[0004] Based on the above problems, the present application provides a method for training a die layout inversion model and a method for determining a die layout, aiming to accurately identify the die layout, so as to improve the pertinence of die layout optimization and improve the thermal management effect.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] In a first aspect, the present application provides a method for training a die layout inversion model, and the method includes:

[0007] Performing a first model training on a preset die layout inversion model by using a simulation data set to obtain a preliminary die layout inversion model; the simulation data set includes simulation data corresponding to multiple groups of heterogeneous integration simulation systems generated by simulation;

[0008] Performing a second model training on the preliminary die layout inversion model by using an experimental data set to obtain a trained die layout inversion model; the experimental data set includes experimental data corresponding to multiple groups of actually constructed heterogeneous integration systems.

[0009] Optionally, in the method as described above, the performing a first model training on a preset die layout inversion model by using a simulation data set to obtain a preliminary die layout inversion model includes:

[0010] Inputting the temperature simulation data in the simulation data set into the preset die layout inversion model to obtain die layout inversion data corresponding to the temperature simulation data; the preset die layout inversion model includes first network model parameters;

[0011] Based on the die layout inversion data corresponding to the temperature simulation data in the simulation dataset, parameter optimization processing is performed on the first network model parameters to obtain second network model parameters;

[0012] Based on the second network model parameters, a preliminary die layout inversion model is determined.

[0013] Optionally, in the method as described above, the first model training is an iterative process, and the first model training of the preset die layout inversion model using the simulation dataset to obtain a preliminary die layout inversion model includes:

[0014] Using the inversion die layout data corresponding to the temperature simulation data obtained by training the preset die layout inversion model in the k-th iteration, calculate the loss value corresponding to the k-th iteration training; k is a positive integer;

[0015] Process the loss value corresponding to the k-th iteration training through a stochastic gradient descent iterative algorithm to obtain the weight gradient of the k-th iteration;

[0016] Based on the network model parameters corresponding to the preset die layout inversion model in the k-th iteration training and the weight gradient of the k-th iteration, determine the network model parameters corresponding to the (k + 1)-th iteration training of the preset die layout inversion model;

[0017] The determining the preliminary die layout inversion model based on the second network model parameters includes:

[0018] Based on the network model parameters corresponding to the preset die layout inversion model in the (k + 1)-th iteration training, determine a preliminary die layout inversion model.

[0019] Optionally, in the method as described above, the inputting the temperature simulation data in the simulation dataset into the preset die layout inversion model to obtain the die layout inversion data corresponding to the temperature simulation data includes:

[0020] Input the temperature simulation data in the simulation dataset into the preset die layout inversion model;

[0021] The temperature simulation data passes through multiple feature extraction layers of the preset die layout inversion model to determine the temperature distribution features corresponding to the temperature simulation data;

[0022] The temperature distribution features corresponding to the temperature simulation data pass through the fully connected layer of the preset die layout inversion model to determine the die layout inversion data corresponding to the temperature simulation data.

[0023] Optionally, in the method as described above, the method for obtaining the simulation dataset includes:

[0024] Build the architecture of the heterogeneous integration simulation system using a thermal simulation tool;

[0025] Based on the preset model parameters, material parameters, and boundary conditions, set the thermal simulation parameters for the architecture of the heterogeneous integration simulation system to obtain the architecture model of the heterogeneous integration simulation system;

[0026] Based on each set of die layout data in the preset multiple sets of die layout data, perform die layout setting on the architecture model of the heterogeneous integration simulation system to obtain multiple heterogeneous integration simulation systems; one set of die layout data corresponds to one heterogeneous integration simulation system; the die layout data includes the number of dies, die power consumption, and die position coordinates;

[0027] Perform thermal simulation processing on each heterogeneous integration simulation system in the multiple heterogeneous integration simulation systems respectively through the thermal simulation tool to generate the temperature simulation data corresponding to each heterogeneous integration simulation system;

[0028] Construct a simulation data set based on the temperature simulation data and die layout data corresponding to each heterogeneous integration simulation system.

[0029] Optionally, for the method as described above, the second model training of the preliminary die layout inversion model using the experimental data set to obtain the trained die layout inversion model includes:

[0030] The second model training is an iterative process. Use the experimental data set to perform m - iteration model training on the preliminary die layout inversion model to obtain the network model parameters corresponding to the (m + 1)-th iteration training preliminary die layout inversion model; m is a positive integer;

[0031] Based on the network model parameters corresponding to the (m + 1)-th iteration training preliminary die layout inversion model, obtain the trained die layout inversion model.

[0032] In a second aspect, the present application provides a method for determining die layout, the method including:

[0033] Obtain the temperature distribution data of the target heterogeneous integration system; the temperature distribution data characterizes the temperature distribution of the heterogeneous integration system at the same moment;

[0034] Use the trained die layout inversion model to perform inversion analysis on the temperature distribution data to determine the target die layout data of the target heterogeneous integration system, so that the user can perform layout optimization processing on the target heterogeneous integration system according to the target die layout data.

[0035] In a third aspect, the present application provides a training device for a die layout inversion model, including:

[0036] A first training module, configured to perform a first model training on a preset die layout inversion model by using a simulation data set, so as to obtain a preliminary die layout inversion model; the simulation data set includes simulation data corresponding to multiple groups of heterogeneous integrated simulation systems generated by simulation.

[0037] A second training module, configured to perform a second model training on the preliminary die layout inversion model by using an experimental data set, so as to obtain a trained die layout inversion model; the experimental data set includes experimental data corresponding to multiple groups of actually constructed heterogeneous integrated systems.

[0038] Fourthly, the present application provides a die layout determination device, including:

[0039] An acquisition module, configured to acquire temperature distribution data of a target heterogeneous integrated system; the temperature distribution data characterizes the temperature distribution of the heterogeneous integrated system at the same moment.

[0040] A processing module, configured to perform inversion analysis on the temperature distribution data by using the trained die layout inversion model, so as to determine target die layout data of the target heterogeneous integrated system, so that a user can perform layout optimization processing on the target heterogeneous integrated system according to the target die layout data.

[0041] Fifthly, the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0042] The memory stores computer-executable instructions;

[0043] The processor executes the computer-executable instructions stored in the memory to implement the training method of the die layout inversion model and the die layout determination method in any one of the above embodiments.

[0044] Sixthly, the present application provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, they are used to implement the training method of the die layout inversion model and the die layout determination method in any one of the above embodiments.

[0045] Compared with the prior art, the present application has the following beneficial effects:

[0046] The preset die layout inversion model is trained for the first time using a simulation data set, and then trained for the second time using an experimental data set. The die layout inversion model trained twice can learn more diverse and comprehensive data features, improving the mapping from temperature distribution data to die layout to be more accurate. Furthermore, by using the trained die layout inversion model above to analyze the temperature distribution data of the target heterogeneous integration system, a more accurate die layout can be deduced, enabling the user to perform layout optimization processing on the target heterogeneous integration system through the accurately deduced die layout data. This helps improve the thermal management effect of the heterogeneous integration system, reduce the hot spot temperature, and enhance the overall performance and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 It is a schematic flowchart of an embodiment of a method for training a die layout inversion model provided by the present application;

[0049] Figure 2 It is a schematic diagram of the architecture of a heterogeneous integration simulation system provided by the present application;

[0050] Figure 3 It is a schematic diagram of the model architecture of a die layout inversion model provided by the present application;

[0051] Figure 4 It is a schematic flowchart of an embodiment of a method for determining a die layout provided by the present application;

[0052] Figure 5 It is a schematic structural diagram of an embodiment of a device for training a die layout inversion model provided by the present application;

[0053] Figure 6 It is a schematic structural diagram of an embodiment of a device for determining a die layout provided by the present application;

[0054] Figure 7 It is a schematic structural diagram of an embodiment of an electronic device provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] To make the objectives, technical solutions, and advantages of this application more clear, the following further elaborates on this application in detail with reference to specific embodiments and the accompanying drawings. It should be noted that the embodiments described in the embodiments of this application are only a part of the embodiments of this application, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts fall within the scope of protection of this application.

[0056] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the embodiments of this application should have the ordinary meaning understood by those with ordinary skills in the field to which this application belongs. The "first", "second", and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are only used to distinguish different components. Words such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. Words such as "connected" or "linked" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right", etc. are only used to represent relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0057] As described above, the existing corelet layout optimization and thermal management technologies mainly rely on the design documents and specifications of corelets. However, the basic information related to corelets recorded is often relatively limited and cannot comprehensively reflect the thermal behavior of corelets during actual operation. Estimation based on limited basic information often fails to accurately obtain the heat source location and power distribution of corelets. This estimation method may ignore factors such as complex heat conduction paths and thermal resistances inside corelets, resulting in a large deviation between the estimation result and the actual situation, and thus lacking pertinence in optimizing the corelet layout and having a poor thermal management effect.

[0058] After research, the inventors proposed a training method for a corelet layout inversion model and a corelet layout determination method to solve the technical problem in the prior art that the accuracy of estimating corelet layout information such as the approximate heat source location and power distribution of corelets is relatively low, resulting in a lack of pertinence in optimizing the corelet layout and a poor thermal management effect.

[0059] To enable those skilled in the art to better understand the solution of this application, the following will clearly and completely describe the technical solution in the embodiments of this application in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

[0060] See Figure 1 , which is a schematic flowchart of an embodiment of a method for training a die layout inversion model provided by this application. As Figure 1 shown, the method includes:

[0061] S101: Use the simulation data set to perform the first model training on the preset die layout inversion model to obtain a preliminary die layout inversion model.

[0062] Among them, the simulation data set contains simulation data corresponding to multiple groups of heterogeneous integrated simulation systems generated by simulation.

[0063] In this embodiment, for example, the function from the initial input to the network output of the preset die layout inversion model can be expressed as S = N(T, w); where T represents the input of the model, N represents the convolutional neural network, S represents the output of the model, and w is the preset network model parameter. Use the simulation data set to perform the first model training on the preset die layout inversion model; obtain a preliminary die layout inversion model, and the function from the input to the network output of this model can be expressed as S opt = N opt (T opt , w opt ); where w opt represents the network model parameter after the first model training; T opt is the temperature simulation data used as the input for the first model training; S opt is the die layout inversion data output by forward propagation using the simulation data set.

[0064] Specifically, the method for obtaining the simulation data set may include:

[0065] Build the architecture of the heterogeneous integrated simulation system using a thermal simulation tool.

[0066] In this embodiment, since the main structure of the heterogeneous integration system includes dies, micro-bumps, underfill materials, redistribution layers, through-silicon vias, and wafer-level silicon substrates. During the process of constructing the architecture model of the heterogeneous integration simulation system, the fine interconnect structure is equivalently processed, and the anisotropic materials of each layer are equivalently converted into isotropic materials; and each component structure is encapsulated, and the heterogeneous integration system is equivalently transformed into the architecture of the heterogeneous integration simulation system connecting each die on the wafer-level silicon substrate, and the architecture of the heterogeneous integration simulation system is built by using the parametric modeling function in the thermal simulation tool, as Figure 2 shown. Among them, the thermal simulation tool is software used to simulate and analyze the thermal behavior of the system. With the parametric modeling ability of the thermal simulation tool, it helps engineers build the architecture of the heterogeneous integration system and analyze its thermal characteristics.

[0067] Based on the preset model parameters, material parameters, and boundary conditions, the thermal simulation parameters of the architecture of the heterogeneous integration simulation system are set to obtain the architecture model of the heterogeneous integration simulation system.

[0068] In this embodiment, based on the preset model parameters, material parameters, and boundary conditions, the thermal simulation parameters of the architecture of the heterogeneous integration simulation system are set. Among them, these parameters include but are not limited to physical layer parameters, physical property parameters such as the thermal conductivity, specific heat capacity, and density of the material, and the boundary conditions of the simulation system, such as temperature, heat flux density, etc. By setting these parameters, the architecture model of the heterogeneous integration simulation system can be obtained.

[0069] Based on each set of die layout data in the preset multiple sets of die layout data, the die layout of the architecture model of the heterogeneous integration simulation system is set to obtain multiple heterogeneous integration simulation systems.

[0070] Among them, one set of die layout data corresponds to one heterogeneous integration simulation system; the die layout data includes the number of dies, the power consumption of the dies, and the die position coordinates.

[0071] In this embodiment, taking the preset C sets of die layout data as an example, each set of die layout data can be represented as D n ; n ∈ C, and n is a positive integer. Based on each set of die layout data in the C sets of die layout data, the die layout of the architecture model of the heterogeneous integration simulation system is set. By applying these layout data to the architecture model, C heterogeneous integration simulation systems with different die layouts can be obtained. Among them, each set of die layout data includes the preset different number of dies as M, different die power consumption as P i and different die position coordinates as (x i , y i , z0); where, i ∈ M, and i is a positive integer; then each set of die layout data can be represented as D n ={M, P i, (x i , y i , z0)}; For example, when the number of different dielets in the system is 3, i can take 1, 2, and 3 respectively, then P1, P2, and P3 respectively represent the power consumption of these 3 different dielets; (x1, y1, z0), (x2, y2, z0), and (x3, y3, z0) respectively represent the position coordinates of these 3 dielets in the system.

[0072] Thermal simulation processing is respectively performed on each heterogeneous integrated simulation system in multiple heterogeneous integrated simulation systems through a thermal simulation tool to generate temperature simulation data corresponding to each heterogeneous integrated simulation system.

[0073] In this embodiment, thermal simulation processing is respectively performed on each of the C heterogeneous integrated simulation systems through a thermal simulation tool, and the temperature prediction function in the thermal simulation tool is used to simulate heat transfer processes such as heat conduction, convection, and radiation for each heterogeneous integrated simulation system, so as to generate temperature simulation data T corresponding to each heterogeneous integrated simulation system n ; where n represents the nth heterogeneous integrated simulation system.

[0074] Based on the temperature simulation data and dielet layout data corresponding to each heterogeneous integrated simulation system, a simulation data set is constructed.

[0075] In this embodiment, based on the temperature simulation data T n and dielet layout data D n corresponding to each of the C heterogeneous integrated simulation systems, a simulation data set is constructed.

[0076] In this embodiment, a thermal simulation tool is used to build the architecture of a heterogeneous integrated simulation system; the temperature distribution of an actual heterogeneous integrated system is simulated through preset model parameters, material parameters, and boundary conditions, and thermal simulation parameters are set for the architecture of the heterogeneous integrated simulation system to obtain an architecture model of the heterogeneous integrated simulation system; based on each set of dielet layout data in a preset multiple sets of dielet layout data, dielet layout settings are performed on the architecture model of the heterogeneous integrated simulation system to obtain multiple heterogeneous integrated simulation systems; thermal simulation processing is respectively performed on each of the multiple heterogeneous integrated simulation systems through a thermal simulation tool to generate temperature simulation data corresponding to each heterogeneous integrated simulation system; based on the temperature simulation data and dielet layout data corresponding to each heterogeneous integrated simulation system, a rich simulation data set is constructed for training a dielet layout inversion model to improve the accuracy and generalization ability of the model, and provide strong data support for subsequent dielet layout optimization.

[0077] The following introduces an example implementation method for performing the first model training on a preset die layout inversion model using a simulation data set to obtain a preliminary die layout inversion model. The implementation method includes the following steps:

[0078] Input the temperature simulation data in the simulation data set into the preset die layout inversion model to obtain die layout inversion data corresponding to the temperature simulation data.

[0079] Among them, the preset die layout inversion model contains first network model parameters.

[0080] In this embodiment, the temperature simulation data in the simulation data set is input into the preset die layout inversion model. After receiving the temperature simulation data, the model will pass through multiple processing layers inside the preset die layout inversion model and finally output die layout inversion data corresponding to the temperature simulation data. The die layout inversion data is the die layout information inferred by the preset die layout inversion model according to the temperature distribution characteristics.

[0081] Based on the die layout inversion data corresponding to the temperature simulation data in the simulation data set, perform parameter optimization processing on the first network model parameters to obtain second network model parameters.

[0082] Based on the second network model parameters, determine the preliminary die layout inversion model.

[0083] In this embodiment, based on the die layout inversion data corresponding to the temperature simulation data in the simulation data set, calculate the difference between the die layout data and the die layout inversion data, that is, the loss function, and use an optimization algorithm to adjust the network model parameters to minimize the difference between the data inverted by the model and the die layout data; and optimize the network model parameters through multiple iterative trainings to obtain the second network model parameters, so as to determine the preliminary die layout inversion model based on the second network model parameters.

[0084] In this embodiment, the temperature simulation data in the simulation data set is input into the preset die layout inversion model to obtain die layout inversion data corresponding to the temperature simulation data; based on the die layout inversion data corresponding to the temperature simulation data in the simulation data set, perform parameter optimization processing on the first network model parameters to obtain second network model parameters; based on the second network model parameters, determine the preliminary die layout inversion model. By optimizing the network model parameters through the die layout data and the die layout inversion data, the accuracy of model inversion can be improved, and the adaptability of the model to different temperature distributions and die layouts can be enhanced.

[0085] As a specific implementation, if the first model training is an iterative process, then the first model training of the preset die layout inversion model using the simulation data set to obtain the preliminary die layout inversion model includes the following steps:

[0086] Using the inversion die layout data and die layout data corresponding to the temperature simulation data obtained by training the preset die layout inversion model in the k-th iteration, calculate the loss value corresponding to the k-th iteration training. Where k is a positive integer.

[0087] Process the loss value corresponding to the k-th iteration training through the stochastic gradient descent iteration algorithm to obtain the weight gradient of the k-th iteration.

[0088] Based on the network model parameters corresponding to the preset die layout inversion model in the k-th iteration training and the weight gradient of the k-th iteration, determine the network model parameters corresponding to the (k + 1)-th iteration training of the preset die layout inversion model.

[0089] Then, based on the second network model parameters, the specific implementation of determining the preliminary die layout inversion model is as follows:

[0090] Based on the network model parameters corresponding to the preset die layout inversion model in the (k + 1)-th iteration training, determine the preliminary die layout inversion model.

[0091] In this embodiment, taking the function from the initial input to the network output of the preset die layout inversion model as S = N(T, w) as an example, in the k-th iteration process, the n-th group of temperature simulation data T n is used as the input of the model, and the die layout inversion data S corresponding to the n-th group of temperature simulation data is obtained respectively n . Based on the n-th group of temperature simulation data T n corresponding die layout inversion data S n and die layout data D n , the mean square error loss function formula is used for minimization operation, and the loss value is calculated using the following formula:

[0092]

[0093] where V is the simulation data set, |V| represents the size of the simulation data set, D n represents the n-th group of die layout data, N(T n ) represents the die layout inversion data inverted from the n-th group of input temperature simulation data T n , which can also be expressed as S n . w (k) represents the network model parameters in the k-th iteration training process.

[0094] Then, the loss value is processed through the stochastic gradient descent iteration algorithm to obtain the weight gradient in the k-th iteration process, and the weight gradient can be expressed by the following formula:

[0095]

[0096] where Δ represents the simulation data set, and |Δ| represents the size of the simulation data set. represents the nabla operator, and the superscript k represents the k-th iteration.

[0097] Then, based on the network model parameter w corresponding to the preset die layout inversion model trained in the k-th iteration (k) and the weight gradient G in the k-th iteration (k) , the network model parameter w corresponding to the preset die layout inversion model trained in the (k + 1)-th iteration is determined. (k+1) It can be obtained by the following formula:

[0098] w (k+1) = w (k) - λG (k)

[0099] where λ is the neural network learning rate.

[0100] Furthermore, based on the network model parameter corresponding to the preset die layout inversion model trained in the (k + 1)-th iteration, a preliminary die layout inversion model is determined.

[0101] In this embodiment, using the inversion die layout data and the die layout data corresponding to the temperature simulation data obtained by training the preset die layout inversion model in the k-th iteration, the loss value corresponding to the k-th iteration training is calculated; the loss value corresponding to the k-th iteration training is processed through the stochastic gradient descent iteration algorithm to obtain the weight gradient in the k-th iteration, so that the model can gradually approach the optimal solution in each iteration, improving the training accuracy; based on the network model parameter corresponding to the preset die layout inversion model trained in the k-th iteration and the weight gradient in the k-th iteration, the network model parameter corresponding to the preset die layout inversion model trained in the (k + 1)-th iteration is determined; based on the network model parameter corresponding to the preset die layout inversion model trained in the (k + 1)-th iteration, a preliminary die layout inversion model is determined. Through multiple iterations of training, the model can gradually learn the complex relationship between the temperature simulation data and the die layout, thereby enhancing the adaptability to different data distributions; and each iteration is dynamically adjusted based on the current weight gradient and model parameters, enabling the model to flexibly adapt to data changes and further improving the generalization ability.

[0102] Further, an example implementation method of inputting the temperature simulation data in the simulation dataset into a preset die layout inversion model to obtain the die layout inversion data corresponding to the temperature simulation data is introduced below. The implementation method includes the following steps:

[0103] Input the temperature simulation data in the simulation dataset into a preset die layout inversion model.

[0104] In this embodiment, as Figure 3 shown, input the temperature simulation data obtained by performing thermal simulation processing on the heterogeneous integration simulation system through a thermal simulation tool into a preset die layout inversion model.

[0105] The temperature simulation data passes through multiple feature extraction layers of the preset die layout inversion model to determine the temperature distribution features corresponding to the temperature simulation data.

[0106] In this embodiment, after the temperature simulation data is input into the preset die layout inversion model, first, the temperature distribution features in the temperature simulation data are extracted through multiple feature extraction layers of the preset die layout inversion model. Among them, as Figure 3 shown, the feature extraction layer is usually a deep learning network structure composed of a convolutional layer and a pooling layer. The extracted feature temperature distribution features may include high and low points of temperature, temperature gradient, heat conduction path, etc., which can reflect the internal thermal distribution characteristics of the heterogeneous integration simulation system.

[0107] The temperature distribution features corresponding to the temperature simulation data pass through the fully connected layer of the preset die layout inversion model to determine the die layout inversion data corresponding to the temperature simulation data.

[0108] In this embodiment, as Figure 3 shown, the extracted temperature distribution features enter the fully connected layer of the model to integrate and transform the extracted temperature distribution features, and finally output the die layout inversion data corresponding to the temperature simulation data.

[0109] In this embodiment, input the temperature simulation data in the simulation dataset into a preset die layout inversion model; the temperature simulation data passes through multiple feature extraction layers of the preset die layout inversion model to determine the temperature distribution features corresponding to the temperature simulation data; the temperature distribution features corresponding to the temperature simulation data pass through the fully connected layer of the preset die layout inversion model to determine the die layout inversion data corresponding to the temperature simulation data, which can improve the accuracy of die layout inversion; and through the learning of a large amount of temperature simulation data, the general law of temperature distribution under different die layouts can be gradually mastered, enhancing the generalization ability of the model.

[0110] S102: Use the experimental dataset to perform a second model training on the preliminary die layout inversion model to obtain a trained die layout inversion model.

[0111] Among them, the experimental data set contains experimental data corresponding to multiple actually constructed heterogeneous integrated systems.

[0112] In this embodiment, due to the differences in experimental conditions and complex external condition changes, there may be certain differences between the die layout data or temperature distribution data of the heterogeneous integrated system in the actual application process and these data obtained by simulation tools. Therefore, in order to improve the inversion accuracy of the die layout inversion model, the preliminary die layout inversion model is trained for the second time using the experimental data set to obtain a trained die layout inversion model. The function of the input to the network output of this model can be expressed as S exp = N exp (T exp , w exp ); where w exp represents the network model parameters after the second model training; T exp is the actual temperature distribution data collected from the experimental data set used as the input for the second model training; S exp is the die layout inversion data output by forward propagation using the experimental data set.

[0113] The following introduces an example implementation method of using the experimental data set to train the preliminary die layout inversion model for the second time to obtain a trained die layout inversion model. The following steps are included in this implementation method:

[0114] The second model training is an iterative process. The preliminary die layout inversion model is trained iteratively m times using the experimental data set to obtain the network model parameters corresponding to the preliminary die layout inversion model after the (m + 1)-th iterative training; m is a positive integer;

[0115] Based on the network model parameters corresponding to the preliminary die layout inversion model after the (m + 1)-th iterative training, a trained die layout inversion model is obtained.

[0116] In the embodiment of the present application, the iterative training using the experimental data set for the preliminary die layout inversion model to obtain a trained die layout inversion model has a similar implementation principle to the first model training process, which will not be elaborated here.

[0117] In this embodiment, through the iterative training process, the preliminary die layout inversion model is trained multiple times using the experimental data set. Based on the network model parameters corresponding to the preliminary die layout inversion model after the (m + 1)-th iterative training, a trained die layout inversion model is obtained, which improves the accuracy, stability, performance, and flexibility of the model, while ensuring the close correlation between the model and the actual application scenario, and providing verifiability and repeatability.

[0118] In this embodiment, the trained die layout inversion model obtained through two rounds of model training is used to analyze the temperature distribution data of the target heterogeneous integration system, and a more accurate die layout can be deduced. Furthermore, the user can perform layout optimization processing on the target heterogeneous integration system through the accurately deduced die layout data, which helps to improve the thermal management effect of the heterogeneous integration system, reduce the hot spot temperature, and improve the overall performance and reliability.

[0119] See Figure 4 , which is a schematic flowchart of an embodiment of a die layout determination method provided by this application. As Figure 4 shown, the method includes:

[0120] S201: Obtain the temperature distribution data of the target heterogeneous integration system.

[0121] Among them, the temperature distribution data characterizes the temperature distribution of the heterogeneous integration system at the same moment.

[0122] In this embodiment, the target heterogeneous integration system is a heterogeneous integration system to be optimized, and this heterogeneous integration system is designed by the user; the target heterogeneous integration system is measured by devices such as temperature sensors or thermal imagers to obtain the corresponding temperature distribution data. This temperature distribution data is a temperature distribution map used to display the temperature change situation in the target heterogeneous integration system. In the temperature distribution map, different colors can be used to represent different temperature values through color coding. For example, as Figure 3 shown in the temperature distribution data input into the model, the colors in the temperature distribution map include blue, green, yellow, orange, and red in ascending order of the represented temperature values; among them, the blue area represents the area with the lowest temperature, and the red area represents the area with the highest temperature.

[0123] S202: Use the trained die layout inversion model to perform inversion analysis on the temperature distribution data, and determine the target die layout data of the target heterogeneous integration system, so that the user can perform layout optimization processing on the target heterogeneous integration system according to the target die layout data.

[0124] In this embodiment, the temperature distribution map of the target heterogeneous integration system is input into the trained die layout inversion model for inversion analysis, and the target die layout data of the target heterogeneous integration system is output, including the number of dice in the target heterogeneous integration system, the power consumption of the dice, and the position coordinates.

[0125] In this embodiment, the temperature distribution data of the target heterogeneous integration system is obtained, and then the trained die layout inversion model is used to perform inversion analysis on the temperature distribution data, and the target die layout data of the target heterogeneous integration system is accurately inverted, so that the user can perform layout optimization processing on the target heterogeneous integration system according to the target die layout data, avoid hotspot concentration, and thus improve the thermal management effect of the heterogeneous integration system.

[0126] Refer to Figure 5 , which is a schematic structural diagram of an embodiment of a training device for a die layout inversion model provided by this application. As Figure 5 shown by the solid line box in, the device 30 includes a first training module 31 and a second training module 32.

[0127] Among them, the first training module 31 is used to perform the first model training on the preset die layout inversion model by using the simulation data set, and obtain a preliminary die layout inversion model; the simulation data set contains simulation data corresponding to multiple groups of heterogeneous integration simulation systems generated by simulation. The second training module 32 is used to perform the second model training on the preliminary die layout inversion model by using the experimental data set, and obtain a trained die layout inversion model; the experimental data set contains experimental data corresponding to multiple groups of actually constructed heterogeneous integration systems.

[0128] The training device for a die layout inversion model provided by the embodiment of this application can execute the technical solutions shown in the above method embodiments, and its implementation principle and beneficial effects are similar, and will not be elaborated here.

[0129] Further, on the basis of the above embodiment, the first training module 31 is specifically used to input the temperature simulation data in the simulation data set into the preset die layout inversion model to obtain die layout inversion data corresponding to the temperature simulation data; the preset die layout inversion model includes first network model parameters; based on the die layout inversion data corresponding to the temperature simulation data in the simulation data set, perform parameter optimization processing on the first network model parameters to obtain second network model parameters; based on the second network model parameters, determine a preliminary die layout inversion model.

[0130] The training device for a die layout inversion model provided by the embodiment of this application can execute the technical solutions shown in the above method embodiments, and its implementation principle and beneficial effects are similar, and will not be elaborated here.

[0131] Further, on the basis of the above embodiments, if the first model training is an iterative process, when the preset die layout inversion model is trained for the first time using the simulation data set to obtain the preliminary die layout inversion model, the first training module 31 is specifically configured to calculate the loss value corresponding to the k-th iteration training by using the inversion die layout data and the die layout data corresponding to the temperature simulation data obtained by training the preset die layout inversion model in the k-th iteration; k is a positive integer; process the loss value corresponding to the k-th iteration training through a stochastic gradient descent iteration algorithm to obtain the weight gradient of the k-th iteration; determine the network model parameters corresponding to the (k + 1)-th iteration training of the preset die layout inversion model based on the network model parameters corresponding to the k-th iteration training of the preset die layout inversion model and the weight gradient of the k-th iteration; when determining the preliminary die layout inversion model based on the second network model parameters, the first training module 31 is specifically configured to determine the preliminary die layout inversion model based on the network model parameters corresponding to the (k + 1)-th iteration training of the preset die layout inversion model.

[0132] The training device for a die layout inversion model provided by an embodiment of the present application can execute the technical solution shown in the above method embodiment, and its implementation principle and beneficial effects are similar, and will not be elaborated here.

[0133] Further, on the basis of the above embodiments, when inputting the temperature simulation data in the simulation data set into the preset die layout inversion model to obtain the die layout inversion data corresponding to the temperature simulation data, the first training module 31 is specifically configured to input the temperature simulation data in the simulation data set into the preset die layout inversion model; the temperature simulation data passes through multiple feature extraction layers of the preset die layout inversion model to determine the temperature distribution feature corresponding to the temperature simulation data; the temperature distribution feature corresponding to the temperature simulation data passes through the fully connected layer of the preset die layout inversion model to determine the die layout inversion data corresponding to the temperature simulation data.

[0134] The training device for a die layout inversion model provided by an embodiment of the present application can execute the technical solution shown in the above method embodiment, and its implementation principle and beneficial effects are similar, and will not be elaborated here.

[0135] Further, on the basis of the above embodiments, as Figure 5 shown by the dashed box in, the device 30 may further include a simulation data set acquisition module 33.

[0136] The simulation dataset acquisition module 33 is used to build the architecture of a heterogeneous integration simulation system using a thermal simulation tool; based on preset model parameters, material parameters, and boundary conditions, perform thermal simulation parameter settings on the architecture of the heterogeneous integration simulation system to obtain an architecture model of the heterogeneous integration simulation system; based on each set of die layout data in a preset multiple sets of die layout data, perform die layout settings on the architecture model of the heterogeneous integration simulation system to obtain multiple heterogeneous integration simulation systems; one set of die layout data corresponds to one heterogeneous integration simulation system; the die layout data includes the number of dies, die power consumption, and die position coordinates; perform thermal simulation processing on each of the multiple heterogeneous integration simulation systems through the thermal simulation tool to generate temperature simulation data corresponding to each heterogeneous integration simulation system; based on the temperature simulation data corresponding to each heterogeneous integration simulation system and the die layout data, construct a simulation dataset.

[0137] The training device for a die layout inversion model provided by an embodiment of the present application can execute the technical solutions shown in the above method embodiments, and its implementation principle and beneficial effects are similar, which will not be elaborated here.

[0138] Further, on the basis of the above embodiment, the second training module 32 is specifically used for the second model training to be an iterative process, using the experimental dataset to perform m - time iterative model training on the preliminary die layout inversion model to obtain the network model parameters corresponding to the (m + 1)-th iterative training preliminary die layout inversion model; m is a positive integer; based on the network model parameters corresponding to the (m + 1)-th iterative training preliminary die layout inversion model, obtain the trained die layout inversion model.

[0139] The training device for a die layout inversion model provided by an embodiment of the present application can execute the technical solutions shown in the above method embodiments, and its implementation principle and beneficial effects are similar, which will not be elaborated here.

[0140] See Figure 6 , this figure is a schematic structural diagram of an embodiment of a die layout determination device provided by the present application. As Figure 6 shown, the device 40 includes an acquisition module 41 and a processing module 42.

[0141] Among them, the acquisition module 41 is used to acquire the temperature distribution data of the target heterogeneous integration system; the temperature distribution data characterizes the temperature distribution of the heterogeneous integration system at the same moment. The processing module 42 is used to perform inversion analysis on the temperature distribution data using the trained die layout inversion model to determine the target die layout data of the target heterogeneous integration system, so that the user can perform layout optimization processing on the target heterogeneous integration system according to the target die layout data.

[0142] The core die layout determination device provided by the embodiments of the present application can execute the technical solutions shown in the above method embodiments, and its implementation principle and beneficial effects are similar, so details are not described herein again.

[0143] See Figure 7 , which is a schematic structural diagram of an embodiment of an electronic device provided by the embodiments of the present application, including:

[0144] A memory 11 for storing computer programs;

[0145] A processor 12 for implementing the steps of a method for training a core die layout inversion model and a method for determining a core die layout as described in any of the above method embodiments when executing the computer program.

[0146] In this embodiment, the device can be an in-vehicle computer, a PC (Personal Computer), or a terminal device such as a smart phone, a tablet computer, a palm computer, or a portable computer.

[0147] The device may include a memory 11, a processor 12, and a bus 13.

[0148] Among them, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, a hard disk, a multimedia card, a card-type memory (such as an SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. The memory 11 can be an internal storage unit of the device in some embodiments, such as the hard disk of the device. The memory 11 can also be an external storage device of the device in other embodiments, such as a plug-in hard disk equipped on the device, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the memory 11 can also include both the internal storage unit and the external storage device of the device. The memory 11 can be used not only to store application software installed on the device and various types of data, such as program codes for executing the method for training a core die layout inversion model and the method for determining a core die layout, but also to temporarily store data that has been output or will be output. The processor 12 can be a Central Processing Unit (CPU) in some embodiments.

[0149] The processor 12 can be a Central Processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments, for running the program codes stored in the memory 11 or processing data, such as program codes for executing the method for training a core die layout inversion model and the method for determining a core die layout.

[0150] The bus 13 may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0151] Furthermore, the device may further include a network interface 14. Optionally, the network interface 14 may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the device and other electronic devices.

[0152] Optionally, the device may further include a user interface 15. The user interface 15 may include a display, an input unit such as a keyboard. Optionally, the user interface 15 may further include a standard wired interface and a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display the information processed in the device and to display a visual user interface.

[0153] Figure 7 Only the device with components 11-15 is shown. Those skilled in the art can understand that Figure 7 the shown structure does not constitute a limitation on the device, and it may include fewer or more components than shown, or combine some components, or have different component arrangements.

[0154] Based on the same inventive concept, corresponding to the method of any of the above embodiments, the embodiment of the present application further provides a computer-readable storage medium. The computer-readable storage medium stores computer instructions, and the computer instructions are used to cause the computer to execute the training method and the core die layout determination method of the core die layout inversion model as described in any of the above embodiments.

[0155] The computer-readable media of the embodiments of the present application include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassette tapes, magnetic tape magnetic disk storage, or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0156] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the training method and the die layout determination method of the die layout inversion model described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0157] It should be noted that the embodiments in this specification are all described in a progressive manner. The same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for methods, devices, electronic devices, and media, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The methods, devices, electronic devices, and media described above are only illustrative. The units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0158] The above is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A training method for a die layout inversion model, characterized in that The method includes: Performing a first model training on a preset die layout inversion model using a simulation dataset to obtain a preliminary die layout inversion model; the simulation dataset contains simulation data corresponding to multiple groups of heterogeneous integrated simulation systems generated by simulation. Performing a second model training on the preliminary die layout inversion model using an experimental dataset to obtain a trained die layout inversion model; the experimental dataset contains experimental data corresponding to multiple groups of actually constructed heterogeneous integrated systems.

2. The method according to claim 1, wherein The performing a first model training on a preset die layout inversion model using a simulation dataset to obtain a preliminary die layout inversion model includes: Inputting the temperature simulation data in the simulation dataset into the preset die layout inversion model to obtain die layout inversion data corresponding to the temperature simulation data; the preset die layout inversion model contains first network model parameters. Based on the die layout inversion data corresponding to the temperature simulation data in the simulation dataset, performing parameter optimization processing on the first network model parameters to obtain second network model parameters. Based on the second network model parameters, determining a preliminary die layout inversion model.

3. The method according to claim 2, wherein The first model training is an iterative process. The performing a first model training on a preset die layout inversion model using a simulation dataset to obtain a preliminary die layout inversion model includes: Calculating a loss value corresponding to the k-th iterative training using the inversion die layout data corresponding to the temperature simulation data obtained by training the preset die layout inversion model in the k-th iteration; k is a positive integer. Processing the loss value corresponding to the k-th iterative training through a stochastic gradient descent iterative algorithm to obtain a weight gradient for the k-th iteration. Based on the network model parameters corresponding to the preset die layout inversion model in the k-th iterative training and the weight gradient for the k-th iteration, determining the network model parameters corresponding to the (k + 1)-th iterative training of the preset die layout inversion model. The determining a preliminary die layout inversion model based on the second network model parameters includes: Based on the network model parameters corresponding to the (k + 1)-th iterative training of the preset die layout inversion model, determining a preliminary die layout inversion model.

4. The method according to claim 3, wherein The inputting the temperature simulation data in the simulation dataset into the preset die layout inversion model to obtain die layout inversion data corresponding to the temperature simulation data includes: Inputting the temperature simulation data in the simulation dataset into the preset die layout inversion model. The temperature simulation data passes through multiple feature extraction layers of the preset die layout inversion model to determine a temperature distribution feature corresponding to the temperature simulation data. The temperature distribution feature corresponding to the temperature simulation data passes through a fully connected layer of the preset die layout inversion model to determine die layout inversion data corresponding to the temperature simulation data.

5. The method according to any one of claims 1-4, characterized in that, The method for obtaining the simulation dataset includes: Using a thermal simulation tool to build the architecture of a heterogeneous integrated simulation system. Based on preset model parameters, material parameters, and boundary conditions, performing thermal simulation parameter settings on the architecture of the heterogeneous integrated simulation system to obtain an architecture model of the heterogeneous integrated simulation system. Based on each set of die layout data in a preset multiple sets of die layout data, perform die layout settings on the architecture model of the heterogeneous integration simulation system to obtain multiple heterogeneous integration simulation systems; one set of die layout data corresponds to one heterogeneous integration simulation system; the die layout data includes the number of dies, die power consumption, and die position coordinates. Through the thermal simulation tool, perform thermal simulation processing on each of the multiple heterogeneous integration simulation systems respectively to generate temperature simulation data corresponding to each of the heterogeneous integration simulation systems. Based on the temperature simulation data corresponding to each heterogeneous integration simulation system and the die layout data, construct a simulation data set.

6. The method according to claim 1, wherein The second model training of the preliminary die layout inversion model using the experimental data set to obtain a trained die layout inversion model includes: The second model training is an iterative process. Use the experimental data set to perform m - times of iterative model training on the preliminary die layout inversion model to obtain the network model parameters corresponding to the (m + 1)-th iterative training preliminary die layout inversion model; m is a positive integer. Based on the network model parameters corresponding to the (m + 1)-th iterative training preliminary die layout inversion model, obtain a trained die layout inversion model.

7. A method for determining a die layout, characterized in that, The method includes: Obtain the temperature distribution data of the target heterogeneous integration system; the temperature distribution data characterizes the temperature distribution of the heterogeneous integration system at the same moment. Use the trained die layout inversion model to perform inversion analysis on the temperature distribution data to determine the target die layout data of the target heterogeneous integration system, so that the user can perform layout optimization processing on the target heterogeneous integration system according to the target die layout data.

8. A training device for a die layout inversion model, characterized in that, Includes: A first training module, configured to perform first - time model training on a preset die layout inversion model using the simulation data set to obtain a preliminary die layout inversion model; the simulation data set contains simulation data corresponding to multiple sets of heterogeneous integration simulation systems generated by simulation. A second training module, configured to perform second - time model training on the preliminary die layout inversion model using the experimental data set to obtain a trained die layout inversion model; the experimental data set contains experimental data corresponding to multiple sets of actually constructed heterogeneous integration systems.

9. A die layout determination device, characterized in that Includes: An acquisition module, configured to obtain the temperature distribution data of the target heterogeneous integration system; the temperature distribution data characterizes the temperature distribution of the heterogeneous integration system at the same moment. A processing module, configured to perform inversion analysis on the temperature distribution data using the trained die layout inversion model to determine the target die layout data of the target heterogeneous integration system, so that the user can perform layout optimization processing on the target heterogeneous integration system according to the target die layout data.

10. An electronic device, characterized in that, The device includes: a processor, and a memory communicatively connected to the processor. The memory stores computer - executable instructions. The processor executes the computer - executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when executed by a processor, are used to implement the method according to any one of claims 1 to 7.

Images