A thermal simulation method for analyzing thermoelectric refrigeration integrated package system

Abstract

This invention provides a thermal simulation method for analyzing thermoelectric cooling integrated packaging systems. The method includes: decomposing the computational domain into thermoelectric and non-thermoelectric regions based on the physical characteristics of the thermoelectric cooling integrated packaging system; employing a one-dimensional adaptive segmentation technique for the thermoelectric region to simplify the three-dimensional thermoelectric arm into a one-dimensional structure, and approximating the temperature-dependent parameters as constants within each segment to establish a port macromodel for the thermoelectric region; constructing a spatially perceptual neural network for the non-thermoelectric region, using a dual-branch architecture to extract spatial heat flow features and global features, and establishing a port macromodel for the non-thermoelectric region; establishing the connection between the thermoelectric and non-thermoelectric regions through the temperature and heat flow continuity conditions at the ports, and performing iterative solutions; and reconstructing the temperature distribution of the non-thermoelectric region based on the heat flow density at the ports. This invention significantly improves the simulation efficiency of thermoelectric cooling integrated packaging systems while ensuring computational accuracy.

Description

Technical Field

[0001] This invention relates to the field of integrated circuit electrothermal coupling simulation technology, and more specifically, to a thermal simulation method for analyzing thermoelectric cooling integrated packaging systems. Background Technology

[0002] With the rapid development of technologies such as artificial intelligence, cloud computing, and autonomous driving, advanced packaging technology has become a key enabling technology for overcoming bandwidth limitations between memory and integrated circuits. However, packaging hotspots generate extremely high local heat flux densities, posing a severe challenge to thermal management in heterogeneous integration. Thermoelectric coolers, based on the Peltier effect, can provide direct and efficient local cooling, effectively solving this problem. In simulating thermoelectric effects, traditional numerical methods require meshing the three-dimensional structure to achieve spatial discretization of the thermoelectric coupling control equations, resulting in a large set of nonlinear equations at each discrete point in space. Solving this set of equations yields the temperature and potential distribution of the overall structure.

[0003] Existing technical solutions suffer from two main problems: First, in the optimization design of thermoelectric cooling integrated packaging systems, designers often need to adjust the geometric dimensions and material parameters of the thermoelectric arm. Each parameter change requires reconstructing the geometry and solving the equations, leading to a significant consumption of computational resources and severely limiting design efficiency. Second, in the thermal design of such cooling systems, designers primarily focus on the maximum operating temperature of the 3D integrated circuit chip with the heat source, while being less sensitive to the detailed temperature distribution of the heat dissipation components. Therefore, developing efficient thermal simulation methods for 3D integrated circuit packaging systems to meet actual engineering needs is a core step in achieving rapid thermal design optimization. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a thermal simulation method for analyzing thermoelectric refrigeration integrated packaging systems. This method performs region decomposition based on physical characteristics and utilizes one-dimensional adaptive segmentation technology and spatial perception neural networks. It is used in the simulation and optimization design process of thermoelectric refrigeration integrated packaging systems, significantly improving simulation efficiency while maintaining computational accuracy.

[0005] The technical solution of the present invention is as follows:

[0006] A thermal simulation method for analyzing thermoelectric refrigeration integrated packaging systems includes the following steps:

[0007] Based on the physical characteristics of the thermoelectric refrigeration integrated packaging system, the computational domain is decomposed into a thermoelectric region and a non-thermoelectric region.

[0008] A one-dimensional adaptive segmentation technique is used for the thermoelectric region to simplify the three-dimensional thermoelectric arm into a one-dimensional structure, and the temperature-dependent parameters are approximated as constants within each segment to establish a port macro model of the thermoelectric region.

[0009] A spatial perception neural network is constructed for the non-thermoelectric region. A dual-branch architecture is used to extract spatial heat flow features and global features, and a port macro model of the non-thermoelectric region is established.

[0010] The connection between the thermoelectric and non-thermoelectric regions is established by using the temperature and heat flow continuity conditions at the port, and the solution is obtained iteratively.

[0011] The temperature distribution in the non-thermoelectric region is reconstructed based on the heat flux density at the port.

[0012] Preferably, in the step of decomposing the computational domain into a thermoelectric region and a non-thermoelectric region based on the physical characteristics of the thermoelectric cooling integrated packaging system, the thermoelectric region is a region that only contains thermoelectric arms, and the non-thermoelectric region is an integrated circuit packaging structure and heat dissipation structure that do not contain thermoelectric effects.

[0013] Preferably, the step of employing a one-dimensional adaptive segmentation technique for the thermoelectric region to simplify the three-dimensional thermoelectric arm into a one-dimensional structure, and approximating the temperature-dependent parameters as constants within each segment to establish a port macromodel of the thermoelectric region specifically includes:

[0014] Step 1: Simplify the three-dimensional thermoelectric arm of the thermoelectric region into a one-dimensional structure;

[0015] The second step is to divide the one-dimensional model into multiple segments and approximate the temperature-dependent parameters as constants within each segment.

[0016] Step 3: Calculate the heat flux density using the temperature at the given port, and dynamically adjust the number of segments according to the adaptive mesh criterion to achieve a balance between computational accuracy and efficiency;

[0017] Step 4: Based on the relationship between temperature and heat flux density at the port, establish a port macro model of the thermoelectric region.

[0018] Preferably, the temperature-dependent parameters include thermal conductivity, electrical conductivity, and Seebeck coefficient.

[0019] Preferably, in the step of constructing a spatial perception neural network for the non-thermoelectric region, extracting spatial heat flux features and global features using a dual-branch architecture, and establishing a port macro model for the non-thermoelectric region, the dual-branch architecture includes: a first branch, which uses a convolutional branch to extract spatial features of heat flux density distribution; and a second branch, which uses a multilayer perceptron branch to process global features of optimization parameters.

[0020] Preferably, in the step of constructing a spatial sensing neural network for the non-thermoelectric region, extracting spatial heat flow features and global features using a dual-branch architecture, and establishing a port macro model for the non-thermoelectric region, when the material properties, heat source distribution, and geometric topology information within the non-thermoelectric region do not change, the trained spatial sensing neural network can be reused to directly obtain the nonlinear temperature response at the port through the port macro model.

[0021] Preferably, the step of establishing the connection between the thermoelectric region and the non-thermoelectric region through the temperature and heat flow continuity conditions at the port, and performing iterative solutions, specifically includes:

[0022] Step 1: Set the initial port temperature at the interface between the thermoelectric and non-thermoelectric regions;

[0023] Step 2: Calculate the heat flux density using the port macromodel of the thermoelectric region;

[0024] Step 3: Input the above heat flux density into the port macro model of the non-thermoelectric region and update the port temperature;

[0025] Step 4: Repeat steps 2 and 3 until the results converge to obtain the final heat flux density distribution at the port.

[0026] Preferably, the step of reconstructing the temperature distribution of the non-thermoelectric region based on the heat flux density at the port specifically includes:

[0027] Step 1: Transform the port heat flux density in the non-thermoelectric region into a second-type boundary condition at the port surface;

[0028] Step 2: Solve the heat conduction equation in the non-thermoelectric region to obtain the temperature distribution results.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. Based on physical properties, the computational domain is decomposed into thermoelectric and non-thermoelectric regions. Appropriate solution methods are adopted for different regions to significantly improve efficiency while maintaining solution accuracy.

[0031] 2. One-dimensional adaptive piecewise segmentation technology is used in the thermoelectric region to effectively handle nonlinear electrothermal coupling equations with temperature-dependent parameters;

[0032] 3. Construct a spatial sensing neural network in the non-thermoelectric region to capture complex spatial heat flow distribution characteristics and achieve efficient temperature prediction; when the geometry and parameters of the non-thermoelectric region do not change, the spatial sensing neural network can be reused to greatly improve the simulation efficiency in the design optimization process. Attached Figure Description

[0033] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0034] Figure 1 This is a flowchart illustrating the thermal simulation method for the thermoelectric refrigeration integrated packaging system of the present invention.

[0035] Figure 2 This is a schematic diagram of the region decomposition strategy proposed in this invention;

[0036] Figure 3 This is a schematic diagram of the convolutional branch of the spatial awareness neural network proposed in this invention;

[0037] Figure 4 This is a schematic diagram of the multilayer perceptron branch of the spatial sensing neural network proposed in this invention;

[0038] Figure 5 This is a schematic diagram of the dual-branch coupled output architecture proposed in this invention;

[0039] Figure 6 This is a schematic diagram of the structure of an integrated thermoelectric cooling packaging system in a specific embodiment;

[0040] Figure 7 The graph shows the maximum temperature of the core particle as a function of voltage under different air inflow velocities. Detailed Implementation

[0041] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0042] Specifically, this invention provides a thermal simulation method for analyzing thermoelectric refrigeration integrated packaging systems, such as... Figure 1 As shown, the method includes the following steps:

[0043] S1: Based on the physical characteristics of the thermoelectric refrigeration integrated packaging system, the computational domain is decomposed into a thermoelectric region and a non-thermoelectric region.

[0044] Specifically, such as Figure 2As shown, in step S1, the thermoelectric region is the region containing only thermoelectric arms, while the non-thermoelectric region is the integrated circuit packaging structure and heat sink structure without thermoelectric effects, including chips, substrates, thermal interface materials, high-bandwidth memory, application-specific integrated circuits, etc. This region decomposition strategy, based on differences in physical properties, allows for the use of the most suitable solution method for different regions. The material properties, heat source distribution, and geometric topology of the non-thermoelectric region remain unchanged during the simulation process, while the size of the thermoelectric arms and the temperature-dependent material properties of the thermoelectric region can be changed during the simulation process.

[0045] S2: A one-dimensional adaptive segmentation technique is used for the thermoelectric region to simplify the three-dimensional thermoelectric arm into a one-dimensional structure, and the temperature dependence parameter is approximated as a constant in each segment to establish a port macro model of the thermoelectric region.

[0046] Specifically, in step S2, the implementation steps of the one-dimensional adaptive segmentation technique are as follows:

[0047] Considering that thermoelectric arms typically have heat flow in only one direction, with almost no components in the other two directions, it is reasonable to simplify the three-dimensional thermoelectric arm of the thermoelectric region into a one-dimensional structure, thus giving a one-dimensional governing equation:

[0048] (1)

[0049] (2)

[0050] Where J is the current density, E = −∇φ is the electric field strength, φ is the electric potential, κ(T) is the temperature-dependent thermal conductivity, T is the temperature, σ(T) is the temperature-dependent electrical conductivity, and S(T) represents the temperature-dependent Seebeck coefficient.

[0051] The single thermoelectric arm is then divided into N segments. Within each segment, σ(T), κ(T), and S(T) are assumed to be constants, where the temperature T is the average temperature of that segment. Explicit expressions for the temperature and current density of the i-th segment can be obtained from formulas (1) and (2):

[0052] (3)

[0053] (4)

[0054] Where c i and d i These are unknown coefficients. Given the initial temperatures at both ends of the thermoelectric arm, considering both the Peltier and Seebeck effects, the heat conduction equations and current continuity equations for each segment are solved. The material parameters and temperature of each segment are iteratively updated until the current density converges in each segment.

[0055] Finally, the temperature distribution within the segment is checked according to the adaptive grid convergence criterion. If the convergence criterion requirement is not met, the segment is further subdivided into two sub-segments, and the above process is repeated until the convergence criterion is met, and the heat flux density of the thermoelectric region ports (hot end and cold end) is finally obtained.

[0056] S3: Construct a spatial perception neural network for the non-thermoelectric region, and use a dual-branch architecture to extract spatial heat flow features and global features to establish a port macro model for the non-thermoelectric region;

[0057] Specifically, the steps for constructing the spatial awareness neural network in step S3 are as follows:

[0058] This spatially perceptual neural network architecture extracts spatial heat flux distribution and global feature coupling through a dual-branch structure. It assumes there are M thermoelectric arms between the thermoelectric and non-thermoelectric regions, and K optimization parameters for the overall heat dissipation system. One approach is a multilayer perceptron branch, which directly processes the complete input vector through two layers of multilayer perceptrons. ,like Figure 3 As shown, this branch consists of a 2M-dimensional hidden layer and an M-dimensional output layer, both activated by the ReLU function. The resulting feature representation is as follows:

[0059] (5)

[0060] It captures the nonlinear relationship between the input vector and the thermal response.

[0061] Another type is the convolutional branch, which focuses on extracting spatial features from the heat flux density distribution, such as... Figure 4 As shown. W, H, C0, C1, and C2 represent the width, height, and channel dimension of the feature, respectively. Input vector It is reshaped into a two-dimensional heatmap and then processed by a convolutional module to obtain a compact representation:

[0062] (6)

[0063] Here, reshape(q) represents the transformation of a 1-D heat flux density vector into a single-channel 2-D image. The module ConvBlock(∙) consists of two convolutional layers, followed by max pooling and adaptive average pooling.

[0064] The outputs of the two branches are concatenated to form a coupled feature vector:

[0065] (7)

[0066] Here, d represents the dimension of the vector extracted from the CNN. Subsequently, the coupled feature vector z is fed into a fusion module, which consists of two fully connected layers, allowing for multi-branch feature integration, thereby improving the accuracy and versatility of thermal modeling. After the dual-branch outputs are coupled, the network is trained using the mean squared error loss function, ultimately yielding the nonlinear response of heat flux density versus temperature at the port, such as... Figure 5 As shown.

[0067] In step S3, when the material properties, heat source distribution, and geometric topology information in the non-thermoelectric region do not change, the trained spatial perception neural network can be reused to directly obtain the nonlinear temperature response at the port through the port macromodel.

[0068] S4: Establish the connection between the thermoelectric region and the non-thermoelectric region by using the temperature and heat flow continuity conditions at the port, and solve iteratively;

[0069] Specifically, step S4 includes the following steps:

[0070] Step 1: Set the initial port temperature at the interface between the thermoelectric and non-thermoelectric regions;

[0071] Step 2: Calculate the heat flux density using the port macromodel of the thermoelectric region;

[0072] Step 3: Input the above heat flux density into the port macro model of the non-thermoelectric region and update the port temperature;

[0073] Step 4: Repeat steps 2 and 3 until the results converge to obtain the final heat flux density distribution at the port.

[0074] S5: Reconstruct the temperature distribution in the non-thermoelectric region based on the heat flux density at the port;

[0075] Specifically, in step S5, the step of reconstructing the temperature distribution of the non-thermoelectric region is as follows:

[0076] The heat flux density that eventually converges is transformed into a second-type boundary condition at the port surface of the non-thermoelectric region. The heat conduction equation of the non-thermoelectric region is solved numerically to reconstruct the temperature distribution of the region of interest.

[0077] Based on the simulation method of the present invention described above, a specific embodiment is calculated.

[0078] The following is in conjunction with the accompanying drawings, Figure 6 The implementation of the technical solution will be further described in detail using the integrated circuit packaging system shown as an example. Figure 6The packaging system structure of this embodiment is illustrated, including a packaging substrate, C4 interconnects, microbumps, an interposer, a physical layer, an epoxy molding compound, a chipset, a high-bandwidth memory, an application-specific integrated circuit (ASIC), and Cu-Cu hybrid bonding. The thermoelectric arm in the thermoelectric cooler has dimensions of 1mm × 1mm × 1.5mm and is made of Bi. 0.52 Sb 1.48 A total of 128 thermocouples, consisting of Te3 and Bi2Te3, are uniformly distributed between two ceramic substrates. To better reflect real-world conditions, a non-isothermal laminar flow physical field is added to the simulation: the upstream temperature T... ustr =300 K, normal inflow velocity of air u In =0.3 m / s.

[0079] In this embodiment, 2275 cases were generated as a training dataset using commercial software by changing the upstream temperature, air inflow velocity, and applied voltage. The spatial awareness neural network proposed in this invention has a maximum absolute error of 1.462 K and a maximum relative error of 0.429%, which is superior to traditional neural networks. The spatial awareness neural network contains 435,900 parameters, has a training time of approximately 62 seconds, and an average inference time of 0.267 ms per case.

[0080] To further verify the accuracy of the method of the present invention, the curves of the maximum internal temperature of the core particle as a function of voltage under different air inflow velocities were calculated, as follows: Figure 7 As shown, the calculation results of the method of this invention are in excellent agreement with those of the finite element method, fully verifying the effectiveness of the method of this invention. In terms of efficiency, the degrees of freedom in the simulation are 5,940,783. The method of this invention takes 11 seconds, while the traditional finite element method takes 4,775 seconds, achieving a speedup of 434 times. It is worth noting that the speedup effect is more significant as the degrees of freedom increase.

[0081] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A thermal simulation method for analyzing thermoelectric refrigeration integrated packaging systems, characterized in that, The method includes the following steps: Based on the physical characteristics of the thermoelectric refrigeration integrated packaging system, the computational domain is decomposed into a thermoelectric region and a non-thermoelectric region. A one-dimensional adaptive segmentation technique is used for the thermoelectric region to simplify the three-dimensional thermoelectric arm into a one-dimensional structure. Within each segment, the temperature-dependent parameters are approximated as constants, establishing a port macro-model of the thermoelectric region, specifically including: Step 1: Simplify the three-dimensional thermoelectric arm of the thermoelectric region into a one-dimensional structure; The second step is to divide the one-dimensional model into multiple segments and approximate the temperature-dependent parameters as constants within each segment. Step 3: Calculate the heat flux density using the temperature at the given port, and dynamically adjust the number of segments according to the adaptive mesh criterion to achieve a balance between computational accuracy and efficiency; Step 4: Based on the relationship between temperature and heat flux density at the port, establish a macro-model of the thermoelectric region at the port. A spatial perception neural network is constructed for the non-thermoelectric region. A dual-branch architecture is used to extract spatial heat flux features and global features, and a port macro model of the non-thermoelectric region is established. The dual-branch architecture includes: a first branch, which uses a convolutional branch to extract spatial features of heat flux density distribution; and a second branch, which uses a multilayer perceptron branch to process global features of optimization parameters. The connection between the thermoelectric and non-thermoelectric regions is established by using the temperature and heat flow continuity conditions at the port, and an iterative solution is performed, specifically including: Step 1: Set the initial port temperature at the interface between the thermoelectric and non-thermoelectric regions; Step 2: Calculate the heat flux density using the port macromodel of the thermoelectric region; Step 3: Input the above heat flux density into the port macro model of the non-thermoelectric region and update the port temperature; Step 4: Repeat steps 2 and 3 until the results converge to obtain the final heat flux density distribution at the port; The temperature distribution in the non-thermoelectric region is reconstructed based on the heat flux density at the port.

2. The thermal simulation method for analyzing thermoelectric refrigeration integrated packaging systems according to claim 1, characterized in that, In the step of decomposing the computational domain into thermoelectric and non-thermoelectric regions based on the physical characteristics of the thermoelectric cooling integrated packaging system, the thermoelectric region is a region that only contains thermoelectric arms, and the non-thermoelectric region is an integrated circuit packaging structure and heat dissipation structure that do not contain thermoelectric effects.

3. The thermal simulation method for analyzing thermoelectric refrigeration integrated packaging systems according to claim 1, characterized in that, The temperature-dependent parameters include thermal conductivity, electrical conductivity, and Seebeck coefficient.

4. The thermal simulation method for analyzing thermoelectric refrigeration integrated packaging systems according to claim 1, characterized in that, In the step of constructing a spatial sensing neural network for the non-thermoelectric region, using a dual-branch architecture to extract spatial heat flow features and global features, and establishing a port macro model for the non-thermoelectric region, when the material properties, heat source distribution, and geometric topology information within the non-thermoelectric region do not change, the trained spatial sensing neural network can be reused to directly obtain the nonlinear temperature response at the port through the port macro model.

5. The thermal simulation method for analyzing thermoelectric refrigeration integrated packaging systems according to claim 1, characterized in that, The step of reconstructing the temperature distribution of the non-thermoelectric region based on the heat flux density at the port specifically includes: Step 1: Transform the port heat flux density in the non-thermoelectric region into a second-type boundary condition at the port surface; Step 2: Solve the heat conduction equation in the non-thermoelectric region to obtain the temperature distribution results.

Images