Heat flow distribution hot spot identification method and device, electronic equipment and storage medium

Abstract

The invention provides a heat flow distribution hot spot identification method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the mesh generation of a to-be-identified system; and then characteristic parameters and heat flow distribution target values of the grids are obtained. And then, for each grid, processing the characteristic parameters through a pre-constructed heat flow distribution prediction deep neural network model to obtain a heat flow distribution prediction value. And for each grid, calculating to obtain an error value of the heat flow distribution target value and the heat flow distribution predicted value. And finally, if the error value is greater than a preset threshold value, marking the current grid as a heat flow distribution hot spot, and outputting position information and characteristic parameters of the current grid. Therefore, by utilizing the method provided by the invention, the heat flow distribution hot spots in the to-be-identified system can be quickly identified, and the heat flow distribution hot spot information is output. The problem that the heat flow distribution hot spots of the heterogeneous integrated system cannot be quickly and accurately identified in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the field of semiconductor technology, and in particular to a method, device, electronic device and storage medium for identifying hot spots of heat flow distribution. Background Art

[0002] Heterogeneous integration, as an important technical choice to continue Moore's Law in the post-Moore era, has received widespread attention in the integrated circuit industry. This technology divides the traditional system-level chip into multiple single-function or multi-function combination bare chips, and then interconnects them through a substrate in a package to form a complete complex functional chip and system.

[0003] With the increase of stacking power density, heat dissipation becomes a major challenge in chip design, which brings great difficulties to the development of heterogeneous integrated advanced packaging technology. Therefore, heterogeneous integrated advanced packaging thermal analysis technology is particularly important. However, there are currently no fully mature thermal analysis tools available for advanced packaging (such as chip-on-chip systems, on-wafer systems, etc.), which makes it impossible to quickly and accurately identify the heat flow distribution hotspots of heterogeneous integrated systems. Summary of the invention

[0004] In view of this, the present application provides a method, device, electronic device and storage medium for identifying heat flux distribution hotspots to solve the problem in the prior art that heat flux distribution hotspots of heterogeneous integrated systems cannot be quickly and accurately identified.

[0005] To achieve the above objectives, this application provides the following technical solutions:

[0006] The first aspect of the present application discloses a method for identifying a heat flux distribution hotspot, comprising:

[0007] Grid division of the system to be identified;

[0008] Acquire characteristic parameters and heat flux distribution target values ​​of each grid; wherein the characteristic parameters are main parameters affecting the heat flux distribution of the system to be identified;

[0009] For each of the grids, the characteristic parameters are processed by a pre-built deep neural network model for heat flow distribution prediction to obtain a heat flow distribution prediction value; wherein the heat flow distribution prediction deep neural network model includes an input layer, a restricted Boltzmann machine hidden layer and a regression prediction output layer;

[0010] For each of the grids, calculating an error value between the heat flux distribution target value and the heat flux distribution predicted value;

[0011] If the error value is greater than a preset threshold, the current grid is marked as a heat flux distribution hotspot, and the position information and characteristic parameters of the current grid are output.

[0012] Optionally, in the above method, the process of constructing the heat flow distribution prediction deep neural network model includes:

[0013] Constructing an experimental data set for the heat flux distribution prediction deep neural network model;

[0014] Based on the experimental data set, the initial parameters of the system to be identified are subjected to dimensionality reduction processing to obtain characteristic parameters;

[0015] Using the characteristic parameters as input layer parameters of the heat flow distribution prediction deep neural network model;

[0016] Pre-training the restricted Boltzmann machine to determine network parameters of the restricted Boltzmann machine;

[0017] The network parameters are used to optimize the regression parameters of the regression prediction output layer to complete the construction of the heat flow distribution prediction deep neural network model.

[0018] Optionally, in the above method, the experimental data set for constructing the heat flow distribution prediction deep neural network model includes:

[0019] Obtaining an initial experimental data set and missing data for the heat flux distribution prediction deep neural network model;

[0020] Based on the initial experimental data set, the missing data is processed to obtain the experimental data to be supplemented;

[0021] The initial experimental data set is supplemented based on the experimental data to be supplemented to obtain the experimental data set.

[0022] Optionally, in the above method, based on the experimental data set, performing dimensionality reduction processing on the initial parameters of the system to be identified to obtain characteristic parameters includes:

[0023] Based on the experimental data set, calculate the characteristic value corresponding to each parameter in the initial parameters;

[0024] Determining the importance level of each of the parameters according to the size of the characteristic value;

[0025] Based on the importance level, parameters in the initial parameters are screened to obtain characteristic parameters.

[0026] Optionally, the above method, before performing dimensionality reduction processing on the initial parameters of the system to be identified based on the experimental data set to obtain characteristic parameters, further includes:

[0027] The initial parameters and the experimental data set are normalized to obtain normalized data.

[0028] The second aspect of the present application discloses a device for identifying a hot spot of heat flow distribution, comprising:

[0029] A division unit is used to divide the system to be identified into grids;

[0030] An acquisition unit, used to acquire characteristic parameters and heat flux distribution target values ​​of each grid; wherein the characteristic parameters are main parameters affecting the heat flux distribution of the system to be identified;

[0031] A prediction unit, for processing the characteristic parameters for each of the grids through a pre-built deep neural network model for heat flow distribution prediction to obtain a predicted value of heat flow distribution; wherein the deep neural network model for heat flow distribution prediction includes an input layer, a restricted Boltzmann machine hidden layer, and a regression prediction output layer;

[0032] A calculation unit, used for calculating, for each of the grids, an error value between the heat flux distribution target value and the heat flux distribution prediction value;

[0033] The identification unit is used to mark the current grid as a heat flux distribution hotspot if the error value is greater than a preset threshold, and output the position information and characteristic parameters of the current grid.

[0034] Optionally, in the above device, the prediction unit includes:

[0035] A construction subunit is used to construct an experimental data set of the heat flow distribution prediction deep neural network model;

[0036] A first data processing subunit is used to perform dimensionality reduction processing on the initial parameters of the system to be identified based on the experimental data set to obtain characteristic parameters;

[0037] A first parameter determination subunit is used to use the characteristic parameters as input layer parameters of the heat flow distribution prediction deep neural network model;

[0038] A second parameter determination subunit, used for pre-training the restricted Boltzmann machine to determine the network parameters of the restricted Boltzmann machine;

[0039] The optimization subunit is used to optimize the regression parameters of the regression prediction output layer using the network parameters to complete the construction of the heat flow distribution prediction deep neural network model.

[0040] Optionally, in the above-mentioned device, the construction subunit includes:

[0041] An acquisition subunit, used to acquire an initial experimental data set and missing data of the heat flow distribution prediction deep neural network model;

[0042] A second data processing subunit is used to perform missing data processing on the missing data based on the initial experimental data set to obtain experimental data to be supplemented;

[0043] The supplementing subunit is used to supplement the initial experimental data set based on the experimental data to be supplemented to obtain the experimental data set.

[0044] Optionally, in the above device, the first data processing subunit includes:

[0045] A calculation subunit, used for calculating, based on the experimental data set, a characteristic value corresponding to each parameter in the initial parameters;

[0046] A determination subunit, used to determine the importance level of each of the parameters according to the size of the characteristic value;

[0047] The screening subunit is used to screen the parameters in the initial parameters based on the importance level to obtain characteristic parameters.

[0048] Optionally, the above device further includes:

[0049] The data processing unit is used to normalize the initial parameters and the experimental data set to obtain normalized data.

[0050] The third aspect of the present application discloses an electronic device, comprising:

[0051] one or more processors;

[0052] a storage device having one or more programs stored thereon;

[0053] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of the first aspects of the present application.

[0054] A fourth aspect of the present application discloses a computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method described in any one of the first aspects of the present application is implemented.

[0055] It can be seen from the above technical scheme that in a method for identifying a heat flux distribution hotspot provided by the present application, the system to be identified is firstly gridded. Then the characteristic parameters and heat flux distribution target values ​​of each grid are obtained; wherein the characteristic parameters are the main parameters affecting the heat flux distribution of the system to be identified. Then, for each grid, the characteristic parameters are processed by a pre-constructed heat flux distribution prediction deep neural network model to obtain a heat flux distribution prediction value; wherein the heat flux distribution prediction deep neural network model includes an input layer, a restricted Boltzmann machine hidden layer and a regression prediction output layer. For each grid, the error value between the heat flux distribution target value and the heat flux distribution prediction value is calculated. Finally, if the error value is greater than a preset threshold, the current grid is marked as a heat flux distribution hotspot, and the position information and characteristic parameters of the current grid are output. It can be seen from this that, using the method of the present application, the characteristic parameters of each grid of the system to be identified are processed by a heat flux distribution prediction deep neural network model to obtain a heat flux distribution prediction value, and the error value between the heat flux distribution target value and the heat flux distribution prediction value is calculated, so as to quickly mark the grid belonging to the heat flux distribution hotspot and output the grid information. The problem that the existing technology cannot quickly and accurately identify the hot spots of heat flow distribution in heterogeneous integrated systems is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0057] Figure 1 A flow chart of a method for identifying a hot spot of heat flux distribution disclosed in an embodiment of the present application;

[0058] Figure 2 A schematic diagram of the structure of a deep neural network model for predicting heat flow distribution disclosed in another embodiment of the present application;

[0059] Figure 3 A flowchart of an implementation of a process for constructing a deep neural network model for heat flow distribution prediction in step S103 disclosed in another embodiment of the present application;

[0060] Figure 4 A schematic diagram of a device for identifying a hot spot of heat flow distribution disclosed in another embodiment of the present application;

[0061] Figure 5 A schematic diagram of an electronic device disclosed in another embodiment of the present application. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0063] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0064] Furthermore, in this document, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0065] From the background technology, we can know that with the increase of stacking power density, heat dissipation becomes a major challenge in chip design, which brings great difficulties to the development of heterogeneous integrated advanced packaging technology. Therefore, heterogeneous integrated advanced packaging thermal analysis technology is particularly important. However, there are currently no fully mature thermal analysis tools for advanced packaging (such as chip-on-chip systems, on-wafer systems, etc.) available, resulting in the inability to quickly and accurately identify the heat flow distribution hotspots of heterogeneous integrated systems.

[0066] In view of this, the present application provides a method, device, electronic device and storage medium for identifying heat flux distribution hotspots to solve the problem in the prior art that heat flux distribution hotspots of heterogeneous integrated systems cannot be quickly and accurately identified.

[0067] The first aspect of the present application discloses a method for identifying a hot spot of heat flow distribution, such as Figure 1 As shown, specifically including:

[0068] S101, gridding the system to be identified.

[0069] It should be noted that, firstly, the system to be identified is grid-divided according to a preset grid size. The system to be identified is a system that needs to identify hot spots of heat flow distribution. In this embodiment, the system to be identified takes a heterogeneous integrated system as an example.

[0070] S102, obtaining characteristic parameters and heat flux distribution target values ​​of each grid, wherein the characteristic parameters are main parameters that affect the heat flux distribution of the system to be identified.

[0071] It should be noted that for each grid, characteristic parameters and heat flux distribution target values ​​within the grid are obtained, wherein the characteristic parameters are the main parameters affecting the flow distribution of the system to be identified, and the heat flux distribution target value is the standard temperature value under normal conditions.

[0072] S103. For each grid, the characteristic parameters are processed by a pre-built deep neural network model for heat flux distribution prediction to obtain a heat flux distribution prediction value; wherein the deep neural network model for heat flux distribution prediction includes an input layer, a restricted Boltzmann machine hidden layer, and a regression prediction output layer.

[0073] It should be noted that the heat flow distribution prediction deep neural network model is Figure 2 As shown, it includes an input layer, a restricted Boltzmann machine hidden layer and a regression prediction output layer. It can be understood that the network structure can be adjusted according to actual conditions. For example, the number of hidden layer nodes and the number of layers can be adjusted, and the number of output nodes can also be adjusted.

[0074] The trained heat flow distribution prediction deep neural network model can be used for heat flow distribution prediction of heterogeneous integrated systems by forward propagation, so as to obtain the heat flow distribution prediction value Ti of the entire heterogeneous integrated system:

[0075] T i =F(ωh N +B) (1)

[0076] Among them, ω and B are the weight and bias parameters of the regression prediction network, h N is the output of the last hidden layer node, and F is the activation function of the regression prediction network.

[0077] After obtaining the characteristic parameters and heat flux distribution target values ​​of each grid, for each grid, the characteristic parameters are processed through the pre-built heat flux distribution prediction deep neural network model to obtain the heat flux distribution prediction value.

[0078] Optionally, in another embodiment of the present application, an implementation method of the construction process of the heat flow distribution prediction deep neural network model in step S103 is as follows: Figure 3 As shown, specifically including:

[0079] S301. Construct an experimental data set for a deep neural network model for predicting heat flow distribution.

[0080] It should be noted that the experimental data set of the deep neural network model for heat flux distribution prediction is first constructed. In the heat flux simulation process, only by establishing a reasonable experimental data set can the neural network structure be fully trained to avoid problems such as model overfitting, thereby achieving accurate prediction of heat flux distribution.

[0081] Optionally, in another embodiment of the present application, an implementation of step S301 may include:

[0082] Obtain the initial experimental data set and missing data for the deep neural network model for heat flux distribution prediction;

[0083] Based on the initial experimental data set, missing data are processed to obtain the experimental data to be supplemented;

[0084] Based on the experimental data to be supplemented, the initial experimental data set is supplemented to obtain the experimental data set.

[0085] It should be noted that the initial experimental data set and missing data of the deep neural network model for heat flow distribution prediction are first used. For example, the experimental data set has M data, of which M1 are missing data, and the M1 missing data are evenly distributed in the experimental data set (the total number of missing data accounts for less than 20%). Then select one of the missing points O, draw a circle with a radius of r, and the circle contains P normal experimental data. The distance r from P point to point O is i The weight is used to make a weighted average of the measured values ​​at point O. The formula is as follows:

[0086]

[0087] Among them, T i is the temperature value of the i-th normal test data. Formula (1) is used to solve the remaining missing point data to obtain the experimental data to be supplemented, and then the initial experimental data set is supplemented based on the experimental data to be supplemented, so that the missing experimental data points can be supplemented and enhanced to obtain a relatively complete experimental data set. At the same time, by adjusting the size of r, the experimental data set can be further optimized.

[0088] S302: Based on the experimental data set, the initial parameters of the system to be identified are subjected to dimensionality reduction processing to obtain characteristic parameters.

[0089] It should be noted that after constructing the experimental data set, the initial parameters of the system to be identified are subjected to dimensionality reduction processing, wherein the initial parameters include packaging design structure parameters, material property parameters, specifically heat sink height, heat sink thickness, thermal interface material thickness, integrated chip position and spacing, geometric size parameters, silicon substrate geometric size parameters, thermal conductivity of various materials, specific heat capacity and density of various materials, convection heat transfer coefficient, TSV structure parameters, solder joint structure parameters, heat source position, etc. The characteristic parameters are the main parameters that affect the heat flow distribution of the heterogeneous integrated system. After dimensionality reduction, the characteristic parameters can be obtained, and the characteristic parameters are the main parameters that affect the heat flow distribution of the heterogeneous integrated system. For example, the thermal conductivity and convection heat transfer coefficient have the largest effect on the heat flow distribution of the heterogeneous integrated system, and the thermal conductivity and convection heat transfer coefficient are the characteristic parameters that affect the heat flow distribution of the heterogeneous integrated system.

[0090] Optionally, in another embodiment of the present application, an implementation of step S302 may include:

[0091] Based on the experimental data set, the eigenvalue corresponding to each parameter in the initial parameters is calculated;

[0092] According to the size of the eigenvalue, determine the importance level of each parameter;

[0093] Based on the importance level, the parameters in the initial parameters are screened to obtain the characteristic parameters.

[0094] It should be noted that, assuming that the experimental data set has M experimental data, the feature parameter group corresponding to the M experimental data is D i (i=1,...,M), first calculate the covariance matrix cov(X,Y) of each dimension, where X and Y are one of the feature parameters (can be the same), then the total covariance matrix Co can be expressed as:

[0095] Co=[(cov(X,Y))] (3)

[0096] Furthermore, the eigenvalue l and eigenvector V of the covariance matrix Co are calculated, where CoV = lV. After obtaining the eigenvalues ​​of the covariance matrix Co, the eigenvalues ​​are sorted according to their sizes to determine the importance level of each parameter. For example, if the eigenvalue is greater than the preset threshold, the parameter level is important, and if the eigenvalue is not greater than the preset threshold, the parameter level is unimportant. Then, based on the importance level, the parameters in the initial parameters are screened, and the variables with smaller effects are ignored to obtain the variables that mainly affect the heat flow distribution of the heterogeneous integrated system, and the parameters with important levels are screened out to obtain the characteristic parameters. For example, the eigenvalues ​​corresponding to the thermal conductivity k and the convective heat transfer coefficient h are the largest, and the importance level is important. The eigenvalues ​​corresponding to the other variables are very small, and the importance level is unimportant. Then the thermal conductivity and the convective heat transfer coefficient are the main factors affecting the heat flow distribution of the heterogeneous integrated system. The corresponding eigenvectors are combined into a new matrix V', and the new characteristic parameter expression can be obtained by the following formula:

[0097] D'=DV' (4)

[0098] Among them, D is the original feature parameter group matrix, and D' is the new feature parameter group matrix after dimensionality reduction.

[0099] Optionally, in another embodiment of the present application, before executing step S302, the following may also be included:

[0100] The initial parameters and experimental data sets are normalized to obtain normalized data.

[0101] It should be noted that the initial parameters and experimental data sets are uniformly marked as X and normalized to avoid model prediction errors caused by inconsistent dimensions. The normalization formula is as follows:

[0102]

[0103] Among them, X' is the normalized data, X min is the minimum value corresponding to the characteristic parameter or experimental data, X max is the maximum value corresponding to the characteristic parameter or experimental data.

[0104] S303, using the characteristic parameters as input layer parameters of a deep neural network model for heat flow distribution prediction.

[0105] It should be noted that after the characteristic parameters are obtained, the characteristic parameters are used as input layer parameters of the deep neural network model for heat flow distribution prediction.

[0106] S304: Pre-train the restricted Boltzmann machine to determine the network parameters of the restricted Boltzmann machine.

[0107] It should be noted that the restricted Boltzmann machine is introduced into the deep neural network model for heat flow distribution prediction mainly to extract high-order features of the deep network input data, so as to establish a better feature expression for model regression prediction. Since the output of the restricted Boltzmann machine is a binary function, in order to process continuous heat flow distribution data, a Gaussian random variable N(0,1) can be added to the visible layer and hidden layer neuron nodes. Therefore, the activation probability of the neurons in the hidden layer and the visible layer can be expressed as:

[0108]

[0109]

[0110] Among them, m and n are the number of nodes in the visible layer and hidden layer, η is a constant value between 0 and 1, and the activation function σ(x) can be expressed as:

[0111]

[0112] σ L and σ H are the lower and upper bounds of the neuron probability, respectively, k 0 is the noise control parameter of the activation function, which is related to the state of the neuron node.

[0113] It should also be noted that in order to obtain the high-dimensional feature expression of the input data (k, h, etc.), the weight parameter w of the continuous data restricted Boltzmann machine needs to be ij , bias parameter c i ,b j And the noise control parameter k 0 The network parameters are iteratively solved, and the contrast divergence algorithm is used to pre-train the Boltzmann machine. The specific process is as follows:

[0114] a. Randomly select training samples and calculate the activation probability h of the hidden layer neurons according to equations (6) and (8) i , and then calculate the activation probability v of the visual layer neurons according to equations (7) and (8) j ;

[0115] b. Update weight parameters, bias parameters, and noise control parameters according to the following formula:

[0116] w ij =λ( <v j ,h i >- <v j ,h i >') (9)

[0117] c i =λ(h i -h i') (10)

[0118] b j =λ(v j -v j ') (11)

[0119] κ j =λ( <v j ,v j >- <v j ,v j >') (12)

[0120] Among them, λ is the learning rate, <·> is the inner product, and ' represents the updated value of the neuron state;

[0121] c. Repeat the above process until the iterative weight parameters, bias parameters and noise control parameters reach the predetermined accuracy or number of times, and then the network parameters of the restricted Boltzmann machine can be determined.

[0122] S305. Optimize the regression parameters of the regression prediction output layer using the network parameters to complete the construction of the deep neural network model for heat flow distribution prediction.

[0123] It should be noted that after the network parameters of the restricted Boltzmann machine are determined, the output of the hidden layer neuron nodes can be used as the input of the regression prediction neural network. By adjusting the network weights and bias parameters, the minimum mean square error between the predicted value and the experimental value of the heat flow distribution of the heterogeneous integrated system can be obtained:

[0124]

[0125] Among them, Q is the regression network weight coefficient w i The penalty factor is N, which is the number of temperature sample points, T i sim and T i exp They represent the predicted and measured values ​​of the heat flow distribution temperature of the heterogeneous integrated system respectively.

[0126] In addition, the network weight coefficient w i It is obtained through the artificial bee colony algorithm. The specific process is as follows:

[0127] a. First, determine the swarm size SN, the maximum number of iterations and the control parameters through the weight parameters, and randomly initialize the swarm (that is, initialize the weight parameters to be solved) to generate SN initial solution vectors X i =(x i1 ,...,x iL ), x ij is the weight parameter w i The artificial bee colony algorithm is represented by , where L is the number of weight parameters.

[0128] b. Neighborhood search, update weight parameter x ij' :

[0129]

[0130] Where s takes values ​​between 1,...,SN, and s is not equal to i. Is a random number between [-1,1]. Calculate the fitness of the updated weights and use the greedy selection strategy to select the best from the new and old weight parameters (if the fitness of the new solution is better than the old solution, remember the new solution and forget the old solution, otherwise keep the old solution). The fitness calculation formula is:

[0131]

[0132] c. Calculate the probability of the food source following the bee:

[0133]

[0134] The follower bee searches for a new possible solution vector according to formula (16). If the probability of the solution is greater than a random number generated in [-1,1], a new solution is generated using formula (14) and its fitness is tested. If it is better than the fitness generated before, the new solution is remembered and the old solution is forgotten. Otherwise, the old solution is retained.

[0135] d. If the solution vector is not updated after a finite number of cycles, the new solution x is calculated by the following formula: ij :

[0136] x ij =x minj +r(x maxj -x minj ) (17)

[0137] Among them, x minj and x maxj They represent the minimum and maximum values ​​of the j-th dimension respectively, and r is a random number between 0 and 1.

[0138] e. Record the optimal solution and determine whether the termination condition is met. If so, output the optimal solution; if not, repeat be until the optimal solution vector is obtained.

[0139] S104. For each grid, calculate the error value between the heat flux distribution target value and the heat flux distribution predicted value.

[0140] It should be noted that for each grid, the error value between the heat flux distribution target value and the heat flux distribution prediction value is calculated, and the calculation formula is as follows:

[0141]

[0142] S105 . If the error value is greater than a preset threshold, the current grid is marked as a heat flux distribution hotspot, and the position information and characteristic parameters of the current grid are output.

[0143] It should be noted that if the error value is greater than a preset threshold, which can be set according to the actual situation, for example, 20%, the current grid is marked as a heat flux distribution hotspot, and the location information and characteristic parameters of the current grid (such as the specific values ​​of k and h) are output. For heat flux distribution hotspots, since the temperature has a large deviation from the target value, the hotspot here is very likely to cause the failure of heterogeneous integrated system chip design and should be avoided as much as possible.

[0144] In a method for identifying a heat flux distribution hotspot provided by the present application, the system to be identified is firstly gridded. Then, the characteristic parameters and heat flux distribution target values ​​of each grid are obtained; wherein, the characteristic parameters are the main parameters affecting the heat flux distribution of the system to be identified. Then, for each grid, the characteristic parameters are processed by a pre-constructed heat flux distribution prediction deep neural network model to obtain a heat flux distribution prediction value; wherein, the heat flux distribution prediction deep neural network model includes an input layer, a restricted Boltzmann machine hidden layer and a regression prediction output layer. For each grid, the error value between the heat flux distribution target value and the heat flux distribution prediction value is calculated. Finally, if the error value is greater than a preset threshold, the current grid is marked as a heat flux distribution hotspot, and the position information and characteristic parameters of the current grid are output. It can be seen from this that, using the method of the present application, the characteristic parameters of each grid of the system to be identified are processed by a heat flux distribution prediction deep neural network model to obtain a heat flux distribution prediction value, and the error value between the heat flux distribution target value and the heat flux distribution prediction value is calculated, so as to quickly mark the grid belonging to the heat flux distribution hotspot and output the grid information. The problem that the heat flux distribution hotspot of the heterogeneous integrated system cannot be quickly and accurately identified in the prior art is solved.

[0145] Another embodiment of the present application also discloses a device for identifying a hot spot of heat flow distribution, such as Figure 4 As shown, specifically including:

[0146] The division unit 401 is used to divide the system to be identified into grids.

[0147] The acquisition unit 402 is used to acquire characteristic parameters and heat flux distribution target values ​​of each grid; wherein the characteristic parameters are main parameters that affect the heat flux distribution of the system to be identified.

[0148] The prediction unit 403 is used to process the characteristic parameters for each grid through a pre-built heat flux distribution prediction deep neural network model to obtain a heat flux distribution prediction value; wherein the heat flux distribution prediction deep neural network model includes an input layer, a restricted Boltzmann machine hidden layer and a regression prediction output layer.

[0149] The calculation unit 404 is used to calculate the error value between the heat flux distribution target value and the heat flux distribution predicted value for each grid.

[0150] The identification unit 405 is used to mark the current grid as a heat flux distribution hotspot if the error value is greater than a preset threshold, and output the position information and characteristic parameters of the current grid.

[0151] In this embodiment, the specific execution process of the division unit 401, the acquisition unit 402, the prediction unit 403, the calculation unit 404, and the identification unit 405 can be found in the corresponding Figure 1 The content of the method embodiment will not be repeated here.

[0152] In a device for identifying a heat flux distribution hotspot provided in the present application, first, a division unit 401 divides the system to be identified into grids. Then, an acquisition unit 402 acquires characteristic parameters and heat flux distribution target values ​​of each grid; wherein, the characteristic parameters are the main parameters affecting the heat flux distribution of the system to be identified. Then, the prediction unit 403 processes the characteristic parameters for each grid through a pre-built heat flux distribution prediction deep neural network model to obtain a heat flux distribution prediction value; wherein, the heat flux distribution prediction deep neural network model includes an input layer, a restricted Boltzmann machine hidden layer, and a regression prediction output layer. The calculation unit 404 calculates the error value between the heat flux distribution target value and the heat flux distribution prediction value for each grid. Finally, if the error value is greater than a preset threshold, the identification unit 405 marks the current grid as a heat flux distribution hotspot and outputs the position information and characteristic parameters of the current grid. It can be seen that, using the method of the present application, the characteristic parameters of each grid of the system to be identified are processed through the heat flow distribution prediction deep neural network model to obtain the heat flow distribution prediction value, and the error value between the heat flow distribution target value and the heat flow distribution prediction value is calculated, so as to quickly mark the grids belonging to the heat flow distribution hotspots and output the grid information. This solves the problem that the existing technology cannot quickly and accurately identify the heat flow distribution hotspots of heterogeneous integrated systems.

[0153] Optionally, in another embodiment of the present application, an implementation of the prediction unit 403 includes:

[0154] Construct subunits for constructing experimental datasets for deep neural network models for heat flux distribution prediction.

[0155] The first data processing subunit is used to perform dimensionality reduction processing on the initial parameters of the system to be identified based on the experimental data set to obtain characteristic parameters.

[0156] The first parameter determination subunit is used to use the characteristic parameters as input layer parameters of the deep neural network model for heat flow distribution prediction.

[0157] The second parameter determination subunit is used to pre-train the restricted Boltzmann machine and determine the network parameters of the restricted Boltzmann machine.

[0158] The optimization subunit is used to optimize the regression parameters of the regression prediction output layer using the network parameters to complete the construction of the deep neural network model for heat flow distribution prediction.

[0159] In this embodiment, the specific execution process of the construction subunit, the first data processing subunit, the first parameter determination subunit, the second parameter determination subunit, and the optimization subunit can be found in the corresponding Figure 3 The content of the method embodiment will not be repeated here.

[0160] Optionally, in another embodiment of the present application, an implementation of the above-mentioned construction subunit includes:

[0161] The acquisition subunit is used to obtain the initial experimental data set and missing data for the deep neural network model for heat flow distribution prediction.

[0162] The second data processing subunit is used to perform missing data processing on the missing data based on the initial experimental data set to obtain the experimental data to be supplemented.

[0163] The supplementing subunit is used to supplement the initial experimental data set based on the experimental data to be supplemented to obtain the experimental data set.

[0164] In this embodiment, the specific execution process of the acquisition subunit, the second data processing subunit, and the supplementation subunit can be found in the above-mentioned corresponding method embodiment content, which will not be repeated here.

[0165] Optionally, in another embodiment of the present application, an implementation of the first data processing subunit includes:

[0166] The calculation subunit is used to calculate the characteristic value corresponding to each parameter in the initial parameters based on the experimental data set.

[0167] The determination subunit is used to determine the importance level of each parameter according to the size of the eigenvalue.

[0168] The screening subunit is used to screen the parameters in the initial parameters based on the importance level to obtain the characteristic parameters.

[0169] In this embodiment, the specific execution process of the calculation subunit, the determination subunit, and the screening subunit can be found in the above-mentioned corresponding method embodiment content, which will not be repeated here.

[0170] Optionally, in another embodiment of the present application, an implementation of the above-mentioned device for identifying hot spots of heat flux distribution includes:

[0171] The data processing unit is used to normalize the initial parameters and the experimental data set to obtain normalized data.

[0172] In this embodiment, the specific execution process of the data processing unit can refer to the corresponding method embodiment content above, which will not be repeated here.

[0173] Another embodiment of the present application further provides an electronic device, such as Figure 5 As shown, specifically including:

[0174] One or more processors 501 .

[0175] The storage device 502 stores one or more programs.

[0176] When one or more programs are executed by one or more processors 501 , the one or more processors 501 implement any one of the methods in the above embodiments.

[0177] Another embodiment of the present application further provides a computer storage medium on which a computer program is stored, wherein when the computer program is executed by a processor, any one of the methods in the above embodiments is implemented.

[0178] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0179] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0180] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying hot spots of heat flux distribution, characterized in that: include: Grid division of the system to be identified; Acquire characteristic parameters and heat flux distribution target values ​​of each grid; wherein the characteristic parameters are main parameters affecting the heat flux distribution of the system to be identified; For each of the grids, the characteristic parameters are processed by a pre-built deep neural network model for heat flow distribution prediction to obtain a heat flow distribution prediction value; wherein the heat flow distribution prediction deep neural network model includes an input layer, a restricted Boltzmann machine hidden layer and a regression prediction output layer; For each of the grids, calculating an error value between the heat flux distribution target value and the heat flux distribution predicted value; If the error value is greater than a preset threshold, the current grid is marked as a heat flux distribution hotspot, and the position information and characteristic parameters of the current grid are output.

2. The method according to claim 1, characterized in that The construction process of the heat flow distribution prediction deep neural network model includes: Constructing an experimental data set for the heat flux distribution prediction deep neural network model; Based on the experimental data set, the initial parameters of the system to be identified are subjected to dimensionality reduction processing to obtain characteristic parameters; Using the characteristic parameters as input layer parameters of the heat flow distribution prediction deep neural network model; Pre-training the restricted Boltzmann machine to determine network parameters of the restricted Boltzmann machine; The network parameters are used to optimize the regression parameters of the regression prediction output layer to complete the construction of the heat flow distribution prediction deep neural network model.

3. The method according to claim 2, characterized in that The experimental data set for constructing the heat flow distribution prediction deep neural network model includes: Obtaining an initial experimental data set and missing data for the heat flux distribution prediction deep neural network model; Based on the initial experimental data set, the missing data is processed to obtain the experimental data to be supplemented; The initial experimental data set is supplemented based on the experimental data to be supplemented to obtain the experimental data set.

4. The method according to claim 2, characterized in that: The step of performing dimensionality reduction processing on the initial parameters of the system to be identified based on the experimental data set to obtain characteristic parameters includes: Based on the experimental data set, calculate the characteristic value corresponding to each parameter in the initial parameters; Determining the importance level of each of the parameters according to the size of the characteristic value; Based on the importance level, parameters in the initial parameters are screened to obtain characteristic parameters.

5. The method according to claim 2, characterized in that: Before the initial parameters of the system to be identified are subjected to dimensionality reduction processing based on the experimental data set to obtain characteristic parameters, the method further includes: The initial parameters and the experimental data set are normalized to obtain normalized data.

6. A device for identifying hot spots of heat flux distribution, characterized in that: include: A division unit is used to divide the system to be identified into grids; An acquisition unit, used to acquire characteristic parameters and heat flux distribution target values ​​of each grid; wherein the characteristic parameters are main parameters affecting the heat flux distribution of the system to be identified; A prediction unit, for processing the characteristic parameters for each of the grids through a pre-built deep neural network model for heat flow distribution prediction to obtain a predicted value of heat flow distribution; wherein the deep neural network model for heat flow distribution prediction includes an input layer, a restricted Boltzmann machine hidden layer, and a regression prediction output layer; A calculation unit, used for calculating, for each of the grids, an error value between the heat flux distribution target value and the heat flux distribution prediction value; The identification unit is used to mark the current grid as a heat flux distribution hotspot if the error value is greater than a preset threshold, and output the position information and characteristic parameters of the current grid.

7. The device according to claim 6, characterized in that The prediction unit comprises: A construction subunit is used to construct an experimental data set of the heat flow distribution prediction deep neural network model; A first data processing subunit is used to perform dimensionality reduction processing on the initial parameters of the system to be identified based on the experimental data set to obtain characteristic parameters; A first parameter determination subunit is used to use the characteristic parameters as input layer parameters of the heat flow distribution prediction deep neural network model; A second parameter determination subunit, used for pre-training the restricted Boltzmann machine to determine the network parameters of the restricted Boltzmann machine; The optimization subunit is used to optimize the regression parameters of the regression prediction output layer using the network parameters to complete the construction of the heat flow distribution prediction deep neural network model.

8. The device according to claim 7, characterized in that The construction subunit comprises: An acquisition subunit, used to acquire an initial experimental data set and missing data of the heat flow distribution prediction deep neural network model; A second data processing subunit is used to perform missing data processing on the missing data based on the initial experimental data set to obtain experimental data to be supplemented; The supplementing subunit is used to supplement the initial experimental data set based on the experimental data to be supplemented to obtain the experimental data set.

9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method according to any one of claims 1 to 5.

10. A computer storage medium, characterized in that: A computer program is stored thereon, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images