Method and device for solving stress of TSV (Through Silicon Via) array
By splitting the TSV array structure model into a subset, using finite element and ANN technology for stress analysis, the accuracy and efficiency problems of TSV array stress analysis in the existing technology are solved, and low-cost and high-accuracy stress solution is achieved.
Patent Information
- Application Number
- CN202411645030.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-03-11
AI Technical Summary
The prior art is difficult to accurately analyze the stress superposition problems caused by multiple TSV and TSV arrays, and the finite element simulation method consumes time and resources and is difficult to ensure the accuracy of the analysis.
By establishing the structural model of the TSV array and splitting it into multiple subsets, using finite element solutions to obtain the stress data of the subset, and then building the model to be trained based on artificial neural network (ANN), training the model through backpropagation algorithm to obtain the full-field stress solution model.
It realizes low-cost and accurate analysis of the stress of the TSV array, and can effectively deal with the stress superposition caused by multiple TSV and TSV arrays to ensure the accuracy of stress analysis.
Smart Images
Figure CN119598916A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of three-dimensional integrated circuits, and in particular to a method and device for solving stress of a TSV array. Background Art
[0002] As the scale of integrated chips continues to expand, three-dimensional integration technology is widely used because it can reduce size, improve performance, and reduce power consumption. Through Silicon Via (TSV) is a key component in three-dimensional integrated circuits. Its preparation process involves a variety of materials. There are also differences in thermal expansion coefficients between materials, which will cause thermal stress in TSV when the temperature rises during operation.
[0003] In the prior art, the stress calculation of TSV is realized through finite element simulation, but the stress superposition caused by multiple TSVs and TSV arrays is relatively complex and requires higher computing power. The finite element simulation method is difficult to ensure the accuracy of stress analysis of multiple TSVs. Therefore, how to accurately realize the stress analysis of TSV arrays has become an urgent problem to be solved. Summary of the invention
[0004] Based on the above problems, the present application provides a method and device for solving TSV array stress, which can accurately and cost-effectively analyze the stress of the TSV array.
[0005] The present application discloses a method for solving stress of a TSV array, the method comprising:
[0006] Establishing a structural model of a TSV array, and splitting the structural model into a plurality of subsets;
[0007] Performing finite element solution based on the subset to obtain stress data of the subset;
[0008] Using the stress data as input data and the full-field stress data of the structural model as output data, a model to be trained is established based on an artificial neural network ANN;
[0009] The model to be trained is trained by a back propagation algorithm, and when the training result meets the preset requirements, a full-field stress solution model is obtained to solve the stress of the TSV array.
[0010] Optionally, taking the stress data as input data includes:
[0011] The stress data is divided into a training set, a verification set and a test set; the training set, the verification set and the test set are disjoint;
[0012] Using the training set as input data to train the model;
[0013] Using the verification set as input data to determine whether to end training;
[0014] The test set is used as input data to evaluate the model performance.
[0015] Optionally, the training the model to be trained by using a back propagation algorithm includes:
[0016] Taking the training set as target data, inputting the geometric parameters and material parameters of the subset by forward propagation;
[0017] The prediction data is calculated through multi-level weighted summation matrix;
[0018] When the difference between the predicted data and the target data does not meet the error requirement, the back propagation algorithm is used to adjust the geometric parameters and the material parameters, and the predicted data is repeatedly calculated until the difference meets the error requirement.
[0019] Optionally, when the training result meets the preset requirements, a full-field stress solution model is obtained, including:
[0020] Input the verification set to obtain the convergence status of the model to be trained;
[0021] When the convergence condition does not meet the preset requirements, repeat the iterative training until the convergence condition meets the preset requirements, and obtain the to-be-trained model as the to-be-tested model;
[0022] Inputting the test set into the model to be tested, obtaining and evaluating the accuracy of the full-field stress data output by the model to be tested;
[0023] When the accuracy does not meet the preset requirement, the iterative training is repeated until the accuracy meets the preset requirement, and the model to be tested is used as the full-field stress solution model.
[0024] Optionally, performing finite element solution based on the subset includes:
[0025] Dividing the subset into finite element grids to obtain a plurality of finite elements;
[0026] There are multiple nodes in the finite element, and the stress of each subset is calculated through the nodes; the nodes correspond to the training temperature, TSV diameter and TSV spacing respectively.
[0027] Based on the above-mentioned method for solving TSV array stress, the present application also discloses a device for solving TSV array stress, including: a splitting unit, a solving unit, a modeling unit and a training unit;
[0028] The splitting unit is used to establish a structural model of the TSV array and split the structural model into multiple subsets;
[0029] The solving unit is used to perform finite element solution based on the subset to obtain stress data of the subset;
[0030] The modeling unit is used to establish a model to be trained based on an artificial neural network ANN, using the stress data as input data and the full-field stress data of the structural model as output data;
[0031] The training unit is used to train the model to be trained by a back propagation algorithm. When the training result meets the preset requirements, a full-field stress solution model is obtained to solve the stress of the TSV array.
[0032] Optionally, the modeling unit includes:
[0033] A dividing subunit, used for dividing the stress data into a training set, a verification set and a test set; the training set, the verification set and the test set are disjoint;
[0034] A training subunit, used for taking the training set as input data for training a model;
[0035] A judgment subunit, used to use the verification set as input data to judge whether to end the training;
[0036] The evaluation subunit is used to use the test set as input data to evaluate model performance.
[0037] Optionally, the training unit includes:
[0038] A parameter input subunit, used to input geometric parameters and material parameters of the subset by forward propagation, taking the training set as target data;
[0039] A calculation subunit, used for calculating the prediction data through a multi-level weighted sum matrix;
[0040] The parameter adjustment subunit is used to adjust the geometric parameters and the material parameters using a back propagation algorithm when the difference value between the predicted data and the target data does not meet the error requirement, and repeatedly calculate the predicted data until the difference value meets the error requirement.
[0041] Optionally, the training unit includes:
[0042] A convergence subunit, used for inputting the verification set to obtain the convergence status of the model to be trained;
[0043] A convergence completion subunit is used to repeat iterative training when the convergence situation does not meet the preset requirements, until the convergence situation meets the preset requirements, and obtain the to-be-trained model as the to-be-tested model;
[0044] A precision subunit, used for inputting the test set into the model to be tested, and obtaining and evaluating the precision of the full-field stress data output by the model to be tested;
[0045] The accuracy completion subunit is used to repeat the iterative training when the accuracy does not meet the preset requirement until the accuracy meets the preset requirement, and use the model to be tested as the full-field stress solution model.
[0046] Optionally, the solving unit includes:
[0047] Dividing sub-units is used to divide the subset into finite element grids to obtain multiple finite elements;
[0048] A solving sub-unit is used for solving a plurality of nodes in the finite element, and the stress of each of the subsets is solved through the nodes; the nodes correspond to the training temperature, TSV diameter and TSV spacing respectively.
[0049] The present application discloses a method and device for solving stress of TSV array, which aims at the problem that finite element simulation is difficult to grasp the full-field stress state of TSV array structure. First, a structural model of TSV array is established and split into multiple subsets. Finite element solution is performed based on the subset, and the stress data of the subset is solved based on the correlation of finite element to load response. The data obtained by finite element simulation is used as input data, and the full-field stress data of the structural model is used as output data, and the real-time solution of the full-field stress field is realized through some stress measurement points. At the same time, a model to be trained is established based on ANN to quickly build model relationships and overcome the time-consuming and resource-consuming shortcomings of finite element analysis. The model to be trained is trained by the back propagation algorithm to obtain the optimal parameters. When the training results meet the preset requirements, a full-field stress solution model is obtained to support the solution of the stress of the TSV array. It can cope with the stress superposition caused by multiple TSVs and TSV arrays to ensure the accuracy of stress analysis. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] 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.
[0051] Figure 1aA schematic flow chart of a method for solving TSV array stress disclosed in an embodiment of the present application;
[0052] Figure 1b A schematic diagram of a 9TSVs structural model disclosed in an embodiment of the present application;
[0053] Figure 1c A schematic diagram of the structure of the model to be trained disclosed in the embodiment of the present application;
[0054] Figure 2 A schematic flow chart of another method for solving TSV array stress disclosed in an embodiment of the present application;
[0055] Figure 3 A schematic diagram of the structure of a device for solving TSV array stress disclosed in an embodiment of the present application. DETAILED DESCRIPTION
[0056] 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.
[0057] Embodiment 1: This application discloses a method for solving TSV array stress.
[0058] For details, please refer to Figure 1a A method for solving stress of a TSV array disclosed in this embodiment includes the following steps:
[0059] Step 101: Establish a structural model of a TSV array, and divide the structural model into multiple subsets.
[0060] In the method described in this embodiment, Figure 1b Schematic diagram of the 9TSVs structural model disclosed in the embodiment of the present application. As an optional method, taking 9TSVs, i.e., a TSV structure with nine through holes, as an example, a 9TSVs structural model is established in the simulation software. Figure 1b As shown in FIG. 1 , the 9TSVs structure model includes nine through-holes penetrating the silicon (Si) substrate, which are filled with conductive materials (such as copper Cu) to form vertical electrical connections. The structural parameters of the model mainly include the diameter of the conductive material, the TSV spacing and the TSV depth. Among them, the TSV depth can be processed as 300μm.
[0061] In the method described in this embodiment, in addition to the structural parameters, some material parameters are also involved in the TSV simulation process. The specific material parameters are shown in the following table:
[0062] Table 1 TSV simulation material parameters
[0063] Material Thermal expansion coefficient (ppm / ℃) Young's modulus(GPa) Poisson's ratio Si 2.3 110 0.35 Cu 17 110 0.35
[0064] In the method described in this embodiment, multiple TSVs are constructed with the above parameters to form a structural model of a TSV array. The structural model is split into several subsets according to a preset splitting basis. For example, a subset may include one TSV, four TSVs, etc. The splitting basis can be set according to demand, and the specific content of the splitting basis is not limited here, and it is sufficient to be able to split the structural model into multiple subsets.
[0065] Step 102: Perform finite element solution based on the subset to obtain stress data of the subset.
[0066] In the method described in this embodiment, the subset obtained by step 101 is divided into finite element grids to obtain multiple finite elements. The responses of these finite elements to the load have similarities and regularities, and the full-field stress of the entire TSV array can be solved by using the stress solution results of a single subset through these similarities and regularities. Among them, the load in the finite element solution process can be a uniform change in temperature, so that the stress cloud map of the subset under different temperature changes can be calculated.
[0067] In the method described in this embodiment, there are multiple nodes in each finite element, and the stress of each subset can be calculated through these nodes. Among them, the nodes can correspond to parameters such as training temperature, TSV diameter and TSV spacing. It should be noted that the results of the finite element solution focus on the stress at the interface between the copper core and the silicon at the center of the TSV.
[0068] As an optional method, the fourth strength theory can be combined to consider the Mises stress (also known as von Mises stress or equivalent stress) value for structural reliability testing. Among them, the fourth strength theory is a theory that describes the yield behavior of materials under complex stress states. Its core content is that when the shape change ratio of a certain point in the material reaches the limit value of the material, plastic yield will occur in the material at that point. Mises stress is used to describe the yield behavior of materials under complex stress states. It assumes that the stress components in all directions are equivalent when the material yields.
[0069] Step 103: Using the stress data as input data and the full-field stress data of the structural model as output data, a model to be trained is established based on an artificial neural network ANN.
[0070] In the method described in this embodiment, as an optional method, the stress data is divided into a non-overlapping training set, a verification set and a test set.
[0071] Artificial neural network (ANN) is a modeling technology that can quickly build a model relationship by training and testing the sampled data obtained by testing or simulation. In order to achieve full-field stress rapid prediction and analysis of TSV array structure, the method described in this embodiment uses ANN as the basis to build a model, thereby solving the time-consuming and resource-consuming problem of finite element analysis of multiple TSVs.
[0072] In the method described in this embodiment, the training set is used to train the model so that the ANN model, i.e., the model to be trained, can have the ability to solve the full-field stress of the TSV array. The verification set is used to determine the ability of the current model to be trained to solve the full-field stress of the TSV array, and to evaluate the convergence of the model to be trained during the training process, and to use this convergence as a basis for determining whether to end the model training. The test set is used to evaluate the performance of the model trained with the training set and the verification set in the task of solving the full-field stress of the TSV array, and to quantify the dispersion of the prediction accuracy of the full-field stress of the TSV array by the trained model.
[0073] Step 104: training the model to be trained by a back propagation algorithm, and when the training result meets the preset requirements, obtaining a full-field stress solution model to solve the stress of the TSV array.
[0074] In the method described in this embodiment, the stress solution result of each finite element in the training set is set as the target data, and the geometric parameters and material parameters of each subset can be input by forward propagation. A multi-level weighted sum matrix is established, and the predicted data is obtained by calculation. Among them, forward propagation refers to the process from the input layer to the output layer of the neural network, and the output of the network is calculated by inputting data and current model parameters (including weights and biases). Its main function is to generate a prediction result (prediction data) and compare it with the true target value (target data) to calculate the loss function (difference value).
[0075] Figure 1c This is a schematic diagram of the structure of the model to be trained disclosed in the embodiment of the present application, such as Figure 1cAs shown in the figure, the first layer of the model to be trained is the input layer, which is responsible for receiving external data (such as training sets, etc.) as input. The middle layer of the model to be trained is the hidden layer, which is responsible for processing and converting the input data. ANN realizes the mapping from input data to output data through the hidden layer. Each neuron in the hidden layer receives the output from the previous layer as input and performs nonlinear transformation through the activation function. The activation function (such as Sigmoid, Tanh, ReLU, etc.) is a key component in the hidden layer, which is responsible for introducing nonlinearity so that the neural network can learn and simulate complex function mappings. Among them, the number of hidden layers and the number of neurons in each hidden layer can be adjusted according to the complexity of the problem. The last layer of the model to be trained is the output layer, which is responsible for generating the final prediction result, that is, the full-field stress of the TSV array.
[0076] In the forward propagation process described in this embodiment, data starts from the input layer, is passed upward layer by layer, and finally reaches the output layer after calculation of the weighted and activation functions of each hidden layer to obtain the predicted output of the model.
[0077] In the neural network training process, forward propagation and back propagation are performed alternately. In the method described in this embodiment, when the difference between the predicted data and the target data does not meet the error requirement, the back propagation algorithm is used to adjust the geometric parameters and material parameters, and the predicted data is repeatedly calculated until the difference meets the error requirement, and finally the matching relationship between the subset and the TSV array stress field is established. Among them, the basis for adjusting the geometric parameters and material parameters can refer to the debugging criteria of the geometric parameters in the actual process. Here, no specific limitation is made on how to adjust the geometric parameters and material parameters. It is sufficient to be able to achieve adjustment and make the difference meet the error requirement.
[0078] In the method described in this embodiment, the convergence of the model to be trained can be obtained by inputting a verification set. The model to be trained has an overfitting phenomenon during the training process. Generally, with the increase of training rounds, the model prediction results show a trend of first becoming excellent, then tending to be flat, and finally decreasing (overfitting). Therefore, it is necessary to monitor its convergence. When the convergence does not meet the preset requirements, iterative training is repeated until the convergence meets the preset requirements, and the model to be trained is obtained as the model to be tested. Among them, the preset requirements can be set according to actual needs, and can be preset values. The content of the preset requirements is not specifically limited here, and it is sufficient to achieve the acquisition of the model to be tested.
[0079] As an optional method, the convergence rule can be set to stop training when the accuracy of the model prediction results obtained by inputting the verification set shows a downward trend and the decrease exceeds a preset value (such as 5%), and obtain the model trained in the previous iteration as the model to be tested.
[0080] In the method described in this embodiment, the test set is input into the model to be tested, and the accuracy of the full-field stress data output by the model to be tested is obtained and further evaluated. When the accuracy does not meet the preset requirements, the iterative training is repeated until the accuracy meets the preset requirements, and the model to be tested is used as the full-field stress solution model. Among them, the preset requirements can be set according to actual needs, and can be preset values. The content of the preset requirements is not specifically limited here, and it is sufficient to achieve the acquisition of the full-field stress solution model. Specifically, it can be to compare the error between the prediction result and the test set. When the error is less than the preset value (such as 0.5%), it is determined that the accuracy meets the preset requirements. It shows that the model to be tested can effectively represent the relationship between temperature load and TSV thermal stress response, and can be used for stress prediction and stress hotspot warning of large structures such as TSV arrays.
[0081] As an optional method, when training, verifying and testing the training model, the specific selection of parameters can be shown in the following table:
[0082] Table 2 Input data definition
[0083]
[0084] Specifically, consider that the input layer consists of 3 nodes, and the corresponding physical quantities are training temperature (T), TSV diameter (D) and TSV spacing (S). Among them, T selects 10 values, D selects 3 values, and S selects 8 values. The final amount of training data is 10×3×8=240, and the number of verification sets and test sets are both 120. Among them, the number of nodes in the input layer, the corresponding physical quantity, the number of physical quantities and other parameters can be set according to actual needs. The amount of data for the training set, verification set and test set can also be set according to demand. When the amount of data for the verification set and the test set is less than that for the training set as described above, certain rules can be set, such as a trigger threshold. And when the input amount of the training set reaches the trigger threshold, start inputting the data for the verification set and the test set to avoid data waste. The above content is not specifically limited here, and it is sufficient to complete the training.
[0085] The method described in this embodiment provides a full-field stress solution method for TSV array based on ANN, which can solve the problem that finite element simulation is difficult to solve the full-field stress of large-size structures, realize the real-time solution of the full-field stress field through subset stress, and realize the real-time visualization of the stress distribution state of the TSV array structure. First, using ANN as the basis, the relationship of the model can be quickly constructed, and the rapid prediction and analysis of the full-field stress of the TSV array structure can be realized, overcoming the time-consuming and resource-consuming shortcomings of finite element analysis. Secondly, based on the correlation between the load responses of finite elements, the full-field stress field of the array is calculated using correlation analysis. At the same time, the model is continuously iterated and optimized through training sets, verification sets and test sets to obtain the optimal parameters. Factors such as different spacing, diameter, temperature, etc. of the TSV array are also taken into account, so that the method proposed in this application can efficiently and accurately coordinate the calculation of the parameters of the TSV array according to different performance requirements, and evaluate the thermal mechanical reliability performance of the TSV array.
[0086] Embodiment 2: This application discloses another method for solving TSV array stress. Figure 2 The method described in this embodiment introduces the whole process of TSV array stress solution.
[0087] Step 201: construct a TSV array model and split it into multiple subsets.
[0088] Step 202: Perform subset stress solution on each subset using the finite element method.
[0089] Step 203: construct a model to be trained based on the ANN, and divide the finite element solution results into a training set, a verification set and a test set.
[0090] Step 204: Input the training set into the model to be trained, and determine whether the output result of the model to be trained meets the first preset requirement. If yes, proceed to step 205. If no, repeat step 204.
[0091] Step 205: Input the verification set into the model to be trained to determine whether the output result of the model to be trained meets the second preset requirement. If yes, use the model to be trained as the model to be tested and proceed to step 206. If no, repeat step 204.
[0092] Step 206: Input the test set into the model to be tested, and determine whether the output result of the model to be tested meets the third preset requirement. If yes, proceed to step 207. If no, repeat step 204.
[0093] Step 207: Acquire the model to be tested as a full-field stress solving model to solve the stress of the TSV array.
[0094] Based on the method for solving TSV array stress disclosed in the above embodiment, this embodiment discloses a device for solving TSV array stress. Figure 3 , the device for solving TSV array stress includes: a splitting unit 301, a solving unit 302, a modeling unit 303 and a training unit 304;
[0095] The splitting unit 301 is used to establish a structural model of the TSV array and split the structural model into multiple subsets;
[0096] The solving unit 302 is used to perform finite element solution based on the subset to obtain stress data of the subset;
[0097] The modeling unit 303 is used to establish a model to be trained based on an artificial neural network ANN, using the stress data as input data and the full-field stress data of the structural model as output data;
[0098] The training unit 304 is used to train the model to be trained by a back propagation algorithm. When the training result meets the preset requirements, a full-field stress solution model is obtained to solve the stress of the TSV array.
[0099] Optionally, the modeling unit 303 includes:
[0100] A dividing subunit, used for dividing the stress data into a training set, a verification set and a test set; the training set, the verification set and the test set are disjoint;
[0101] A training subunit, used for taking the training set as input data for training a model;
[0102] A judgment subunit, used to use the verification set as input data to judge whether to end the training;
[0103] The evaluation subunit is used to use the test set as input data to evaluate model performance.
[0104] Optionally, the training unit 304 includes:
[0105] A parameter input subunit, used to input geometric parameters and material parameters of the subset by forward propagation, taking the training set as target data;
[0106] A calculation subunit, used for calculating the prediction data through a multi-level weighted sum matrix;
[0107] The parameter adjustment subunit is used to adjust the geometric parameters and the material parameters using a back propagation algorithm when the difference value between the predicted data and the target data does not meet the error requirement, and repeatedly calculate the predicted data until the difference value meets the error requirement.
[0108] Optionally, the training unit 304 includes:
[0109] A convergence subunit, used for inputting the verification set to obtain the convergence status of the model to be trained;
[0110] A convergence completion subunit is used to repeat iterative training when the convergence situation does not meet the preset requirements, until the convergence situation meets the preset requirements, and obtain the to-be-trained model as the to-be-tested model;
[0111] A precision subunit, used for inputting the test set into the model to be tested, and obtaining and evaluating the precision of the full-field stress data output by the model to be tested;
[0112] The accuracy completion subunit is used to repeat the iterative training when the accuracy does not meet the preset requirement until the accuracy meets the preset requirement, and use the model to be tested as the full-field stress solution model.
[0113] Optionally, the solving unit 302 includes:
[0114] Dividing sub-units is used to divide the subset into finite element grids to obtain multiple finite elements;
[0115] A solving sub-unit is used for solving a plurality of nodes in the finite element, and the stress of each of the subsets is solved through the nodes; the nodes correspond to the training temperature, TSV diameter and TSV spacing respectively.
[0116] The embodiments in this specification are described in a progressive manner. As for the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the methods.
[0117] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including 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, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0118] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0119] The features described in the embodiments of this specification can be replaced or combined with each other, so that professional and technical personnel in this field can implement or use the present application.
[0120] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. 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 solving stress of a TSV array, characterized in that: include: Establishing a structural model of a TSV array, and splitting the structural model into a plurality of subsets; Performing finite element solution based on the subset to obtain stress data of the subset; Taking the stress data as input data and the full-field stress data of the structural model as output data, a model to be trained is established based on an artificial neural network ANN; The model to be trained is trained by a back propagation algorithm, and when the training result meets the preset requirements, a full-field stress solution model is obtained to solve the stress of the TSV array.
2. The method according to claim 1, characterized in that The method of taking the stress data as input data comprises: The stress data is divided into a training set, a verification set and a test set; the training set, the verification set and the test set are disjoint; Using the training set as input data to train the model; Using the verification set as input data to determine whether to end training; The test set is used as input data to evaluate the model performance.
3. The method according to claim 2, characterized in that The training of the model to be trained by a back propagation algorithm includes: Taking the training set as target data, inputting the geometric parameters and material parameters of the subset by forward propagation; The prediction data is calculated through multi-level weighted summation matrix; When the difference between the predicted data and the target data does not meet the error requirement, the back propagation algorithm is used to adjust the geometric parameters and the material parameters, and the predicted data is repeatedly calculated until the difference meets the error requirement.
4. The method according to claim 2, characterized in that: When the training results meet the preset requirements, a full-field stress solution model is obtained, including: Input the verification set to obtain the convergence status of the model to be trained; When the convergence condition does not meet the preset requirements, repeat the iterative training until the convergence condition meets the preset requirements, and obtain the to-be-trained model as the to-be-tested model; Inputting the test set into the model to be tested, obtaining and evaluating the accuracy of the full-field stress data output by the model to be tested; When the accuracy does not meet the preset requirement, the iterative training is repeated until the accuracy meets the preset requirement, and the model to be tested is used as the full-field stress solution model.
5. The method according to claim 1, characterized in that The performing finite element solution based on the subset includes: Dividing the subset into finite element grids to obtain a plurality of finite elements; There are multiple nodes in the finite element, and the stress of each subset is calculated through the nodes; the nodes correspond to the training temperature, TSV diameter and TSV spacing respectively.
6. A device for solving TSV array stress, characterized in that: include: Splitting unit, solving unit, modeling unit and training unit; The splitting unit is used to establish a structural model of the TSV array and split the structural model into multiple subsets; The solving unit is used to perform finite element solution based on the subset to obtain stress data of the subset; The modeling unit is used to establish a model to be trained based on an artificial neural network ANN, using the stress data as input data and the full-field stress data of the structural model as output data; The training unit is used to train the model to be trained by a back propagation algorithm. When the training result meets the preset requirements, a full-field stress solution model is obtained to solve the stress of the TSV array.
7. The device according to claim 6, characterized in that The modeling unit comprises: A dividing subunit, used for dividing the stress data into a training set, a verification set and a test set; the training set, the verification set and the test set are disjoint; A training subunit, used for taking the training set as input data for training a model; A judgment subunit, used to use the verification set as input data to judge whether to end the training; The evaluation subunit is used to use the test set as input data to evaluate model performance.
8. The device according to claim 7, characterized in that The training unit comprises: A parameter input subunit, used to input the geometric parameters and material parameters of the subset by forward propagation, taking the training set as target data; A calculation subunit, used for calculating the prediction data through a multi-level weighted sum matrix; The parameter adjustment subunit is used to adjust the geometric parameters and the material parameters using a back propagation algorithm when the difference value between the predicted data and the target data does not meet the error requirement, and repeatedly calculate the predicted data until the difference value meets the error requirement.
9. The device according to claim 7, characterized in that The training unit comprises: A convergence subunit, used for inputting the verification set to obtain the convergence status of the model to be trained; A convergence completion subunit is used to repeat iterative training when the convergence situation does not meet the preset requirements, until the convergence situation meets the preset requirements, and obtain the to-be-trained model as the to-be-tested model; A precision subunit, used for inputting the test set into the model to be tested, and obtaining and evaluating the precision of the full-field stress data output by the model to be tested; The accuracy completion subunit is used to repeat the iterative training when the accuracy does not meet the preset requirement until the accuracy meets the preset requirement, and use the model to be tested as the full-field stress solution model.
10. The device according to claim 6, characterized in that The solving unit comprises: Dividing sub-units is used to divide the subset into finite element grids to obtain multiple finite elements; A solving sub-unit is used for solving a plurality of nodes in the finite element, and the stress of each of the subsets is solved through the nodes; the nodes correspond to the training temperature, TSV diameter and TSV spacing respectively.