Accurate heat dissipation micro-channel design method for local hot spots of high-power chip

Abstract

The invention discloses a precise heat dissipation micro-channel design method for local hot spots of a high-power chip, and belongs to the field of chip heat dissipation. According to the method, the structure of the micro-channel radiator is designed according to the heat source distribution and the size specification of the high-power chip, the space between turbulent flow elements is determined based on a hot spot position fitting function, function parameters are iterated repeatedly according to the average temperature, precise heat dissipation of local hot spots is achieved, and the method is suitable for a chip system with the thermal power of kilowatt level.

Description

Technical Field

[0001] The present invention relates to the technical field of chip heat dissipation, and particularly to a precise heat dissipation microchannel design method for local hot spots of high-power chips. Background Art

[0002] With the growth of the global data volume, the high data volume and high computing power in data centers will bring high power consumption problems to heat dissipation devices. Micro heat dissipation components have many advantages such as high heat transfer efficiency and good engineering feasibility. Applying a microchannel radiator and designing an embedded heat conduction component can improve the heat dissipation efficiency. Designing microchannel heat dissipation devices around new AI acceleration chips such as CPUs, GPUs, TPUs, and wafer-level chips can effectively improve the heat dissipation system and enhance energy consumption effectiveness.

[0003] Existing design methods improve the design from aspects such as the size, structure, and material of the microchannel radiator, focusing on increasing the heat transfer area and improving the heat transfer capacity to meet the requirements of high-power radiators. However, such design methods often do not accurately design the turbulence element structure according to the heat source distribution, resulting in poor heat dissipation effect and uneven temperature distribution. In actual application scenarios, a microchannel radiator for precise heat dissipation of local hot spots of high-power chips is desired. Summary of the Invention

[0004] The present invention provides a precise heat dissipation microchannel design method for local hot spots of high-power chips, which can achieve precise heat dissipation of local hot spots of the chips.

[0005] An embodiment of the present invention provides a precise heat dissipation microchannel design method for local hot spots of high-power chips, including the following steps:

[0006] Obtain the heat source distribution and size specifications of the high-power chip, determine the heat source points according to the heat source distribution and size specifications of the high-power chip, and set the initial turbulence element spacing according to the heat source power;

[0007] Fit a parametric binary function in the XY plane based on the initial turbulence element spacing and the heat source points, and determine the optimal turbulence element spacing by iterating the parametric binary function multiple times according to the average temperature of the heat source;

[0008] Determine the structure of the microchannel radiator according to the optimal turbulence element spacing.

[0009] Optionally, in an embodiment of the present invention, determining the optimal turbulence element spacing by iterating the parametric binary function multiple times according to the average temperature of the heat source includes:

[0010] Set the initial values of the control parameters in the two-parameter function and determine the iteration step size. Increase or decrease the control parameters to change the spacing function value, and recalculate the average temperature of the heat source. Determine whether the heat transfer capacity under the current two-parameter function is greater than that of the previous two-parameter function. When it is greater, continue the iteration until the heat transfer capacity is less than that of the previous two-parameter function. Determine the current control parameters and continue with multiple rounds of iteration. If the heat transfer capacity under the current two-parameter function is greater than that of the previous round of two-parameter function, continue the iteration until the heat transfer capacity is less than that of the previous two-parameter function. Determine the current control parameters. If the heat transfer capacity under the current two-parameter function is not greater than that of the previous round of two-parameter function until the maximum number of iterations is reached, output the current control parameter value and determine the optimal spoiler element spacing.

[0011] Optionally, in an embodiment of the present invention, the average temperature of the heat source is calculated according to the heat transfer amount between the liquid and the microchannel heat sink under the microchannel heat sink structure. The heat transfer amount between the liquid and the microchannel heat sink under the microchannel heat sink structure is:

[0012] Q = hAΔT

[0013] where h is the convective heat transfer coefficient, A is the area of the interface between the liquid and the microchannel, and ΔT is a constant.

[0014] Optionally, in an embodiment of the present invention, in determining the structure of the microchannel heat sink according to the optimal spoiler element spacing, the spoiler element is rectangular.

[0015] The precise heat dissipation microchannel design method for local hot spots of high-power chips according to the embodiments of the present invention designs the structure of the microchannel heat sink according to the heat source distribution and size specifications of the high-power chips, determines the spoiler element spacing based on the hot spot position fitting function, and repeatedly iterates the function parameters according to the average temperature to achieve precise heat dissipation of local hot spots, and is applicable to chip systems with a thermal power of kilowatts.

[0016] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be understood through the practice of the present invention. Description of the Drawings

[0017] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the following description of the embodiments in conjunction with the drawings, where:

[0018] Figure 1 is a flowchart of a precise heat dissipation microchannel design method for local hot spots of high-power chips according to an embodiment of the present invention;

[0019] Figure 2 is a schematic diagram of the iteration process of an embodiment of the present invention;

[0020] Figure 3 A schematic diagram of the NVIDIA B200 high-power chip according to an embodiment of the present invention;

[0021] Figure 4 A high-power chip model designed for an embodiment of the present invention;

[0022] Figure 5 It is a schematic plan view of a rectangular spoiler structure according to an embodiment of the present invention;

[0023] Figure 6 A three-dimensional structural diagram of a microchannel heat sink modeled according to an embodiment of the present invention;

[0024] Figure 7 4 is a curve diagram showing the relationship between the average temperature and the control parameters of an embodiment of the present invention. DETAILED DESCRIPTION

[0025] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0026] Figure 1 The present invention provides a flowchart of a method for designing precise heat dissipation microchannels for local hot spots of high-power chips according to an embodiment of the present invention.

[0027] like Figure 1 As shown, the precise heat dissipation microchannel design method for local hot spots of high-power chips includes the following steps:

[0028] Step S1, obtaining the heat source distribution and size specifications of the high-power chip, determining the heat source point according to the heat source distribution and size specifications of the high-power chip, and setting the initial spoiler element spacing according to the heat source power.

[0029] When designing the heat sink structure, it is necessary to consider the heat source distribution and size of the high-power chip and design the heat sink structure in a targeted manner to achieve better heat dissipation effect.

[0030] Step S2, fitting a parameter-containing binary function in the XY plane based on the initial spoiler element spacing and the heat source point position, and iterating the parameter-containing binary function multiple times according to the average temperature of the heat source to determine the optimal spoiler element spacing.

[0031] The control parameter in the two-parameter function is used to iteratively and precisely control the spacing of the turbulators. The average temperature of the heat source is calculated as the basis for iteration. In the embodiments of the present invention, when designing the radiator structure, the influence of different control parameters on the heat dissipation effect is considered. First, the control parameter is set to 1 for calculation. The structure of the microchannel radiator under the current control parameter constructed according to the heat source distribution and size specifications of the high-power chip can be a rectangular turbulator structure or other forms of turbulent structures, which is not limited herein. In the following embodiments, the rectangular turbulator structure is taken as an example for illustration.

[0032] Step S3: Determine the structure of the microchannel radiator according to the optimal turbulator spacing.

[0033] In an embodiment of the present invention, the optimal turbulator spacing is determined by iteratively calculating the two-parameter function based on the average temperature of the heat source, including:

[0034] Set the initial value of the control parameter in the two-parameter function and determine the iteration step size. Increase or decrease the control parameter to change the value of the spacing function, and recalculate the average temperature of the heat source. Determine whether the heat transfer capacity under the current two-parameter function is greater than that of the previous two-parameter function. If it is greater, continue the iteration until the heat transfer capacity is less than that of the previous two-parameter function. Determine the current control parameter and continue multiple rounds of iteration. If the heat transfer capacity under the current two-parameter function is greater than that of the previous round of the two-parameter function, continue the iteration until the heat transfer capacity is less than that of the previous two-parameter function. Determine the current control parameter. If the heat transfer capacity under the current two-parameter function is not greater than that of the previous round of the two-parameter function until the maximum number of iterations is reached, output the current control parameter value and determine the optimal turbulator spacing.

[0035] In the embodiments of the present invention, the heat transfer amount between the liquid and the microchannel radiator under the current microchannel radiator structure is calculated, and the average temperature of the chip surface is calculated to measure the heat dissipation performance of the radiator. The heat transfer amount between the liquid and the microchannel radiator is:

[0036] Q = hAΔT

[0037] where h is the convective heat transfer coefficient, A is the area of the interface between the liquid and the microchannel, and ΔT is a constant.

[0038] In step S3, a fixed step size of 0.1 is taken. The current control parameter is added with this step size, and the average temperature under the current control parameter is recalculated. It is judged whether the heat exchange efficiency under the control parameter of the current round is greater than the heat exchange efficiency under the control parameter of the previous round. If the heat exchange efficiency of the parametric binary function in the current round is greater than the heat exchange efficiency of the parametric binary function in the previous round, continue the next round of iteration until the heat exchange efficiency of the new round of parametric binary function is less than the heat exchange efficiency of the parametric binary function in the previous round. Determine the control parameter K at this time, and continue to increase the step size for a certain number of iterations. If the heat exchange efficiency greater than the heat exchange efficiency corresponding to the control parameter K appears during this iteration process, determine the new control parameter K1, and on the basis of the control parameter K1, continue to increase the step size, and judge whether the heat exchange efficiency of the next round of parametric binary function is greater than the heat exchange efficiency of the parametric binary function in the previous round, repeat the judgment process until the heat exchange efficiency of the next round of parametric binary function is less than the heat exchange efficiency of the parametric binary function in the previous round, and no parametric binary function with higher heat exchange efficiency appears after experiencing the specified number of iterations, then determine the parametric binary function at this time as the optimal parametric binary function, obtain the optimal control parameter, and design the radiator structure. The iteration process is as shown in the appendix Figure 2 as shown.

[0039] For example, in the specific embodiment of the present invention, the initial control parameter k = 1 is taken, and the fixed step size is 0.1. In the new round of iteration, k = 1.1. It is judged that the heat exchange efficiency of the parametric binary function at this time is lower than the heat exchange efficiency of the parametric binary function when k = 1. Then, five iterations are performed between k = 1.2 and 1.7. In the iteration results, the heat exchange efficiency greater than the parametric binary function when k = 1 does not appear in the five iterations. Then, it is determined that k = 1 is the optimal parametric binary function in the positive direction. For the reverse iteration, the step size is taken as -0.1, and the above iteration process is repeated.

[0040] As Figure 3 shown, the B200 GPU is based on TSMC's N4P process technology, and the number of transistors reaches 208 billion, which is more than twice that of 80 billion transistors of H100 / H200. This also enables the artificial intelligence performance of B200 to reach 20 petaflops. B200 uses two chips with full mask sizes, and there are four HMB3e stacks around each die, each stack is 24GB, and each stack has a bandwidth of 1TB / s on a 1024bit interface. However, it also brings challenges in chip heat dissipation, and the thermal design power consumption is as high as 1200W.

[0041] Figure 4 For the designed high-power chip model, the brazing packaging process, the aluminum shell heat conduction layer, and the silicone grease heat conduction layer for heat dissipation requirements are simulated, which fits the real situation to the greatest extent.

[0042] In this model, an XY plane is established with the center as the origin. The CPU die serves as the main heat source, with a range of -24mm < x < 24mm and -18mm < y < 18mm. Based on this main heat source, a spacing function needs to be fitted on the XY plane. The function is required to be denser in the range of -24mm < x < 24mm and -18mm < y < 18mm and looser in other domains to achieve the best heat dissipation effect. A binary polynomial can be selected as a function similar to this.

[0043] The method of the present invention will be described in detail below through a specific embodiment.

[0044] Step 1: Determine the size of the radiator. Both the length and width are 72mm. With the center as the origin, the domain of the function is -36 < x < 36 and -36 < y < 36. The range of the main heat source is -24mm < x < 24mm and -18mm < y < 18mm. Under these conditions, fit the binary function:

[0045]

[0046] where k is a control parameter used to finely adjust the spacing size, and b is the basic spacing.

[0047] Step 2: Set the parameters k = 1 and b = 1.5.

[0048] Step 3: Calculate parameters such as the heat transfer area and the average effective length.

[0049] Step 4: Calculate the average temperature.

[0050] As k decreases, the heat transfer area A becomes denser as the turbulators become denser. According to the Navier - Stokes equations describing the flow of viscous fluids, overly dense turbulators will lead to insufficient heat transfer.

[0051] Let's assume the extreme case of k = 0. At this time, the spacing is 0, that is, a straight - channel radiator. Obviously, its heat dissipation efficiency is much lower than that of the micro - channel design with turbulators.

[0052] In this example, the average temperature first decreases and then increases as the control parameter increases. There is an optimal heat dissipation efficiency when k is near 1.

[0053] Model the rectangular turbulator radiator and perform finite - element simulation calculations. As Figure 5 and Figure 6 shown, it is the three - dimensional structure diagram of the model. The present invention designs an approximate model of the NVIDIA B200 high - power chip and fits a spacing function based on this model to accurately design the test micro - channel radiator.

[0054] The simulation results are as follows:

[0055] Control parameter k Average temperature T / K 0 355.5 0.5 354.3 1 353.9 1.5 354.1 2 354.8

[0056] Figure 7 The relationship curve of T-k is given. By fitting the data points, it is obtained that when k is near 1, T is the smallest, that is, the best heat dissipation efficiency is achieved.

[0057] According to the precise heat dissipation microchannel design method for local hot spots of high-power chips proposed in the embodiments of the present invention, the structure of the microchannel radiator is designed according to the heat source distribution and size specifications of the high-power chip, and the spacing of the flow disturbing elements is determined based on the fitting function of the hot spot position. The function parameters are iteratively repeated according to the average temperature to achieve precise heat dissipation of the local hot spot, which is applicable to chip systems with a thermal power of kilowatts.

[0058] In the description of this specification, the descriptions with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or N embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0059] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" can explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "N" is at least two, such as two, three, etc., unless otherwise specifically defined.

[0060] Any process or method description in the flowchart or described in other ways herein can be understood as representing a module, segment, or part of code including one or N executable instructions for implementing a customized logic function or process. The scope of the preferred embodiments of the present invention includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in the reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

Claims

1. A precise heat dissipation microchannel design method for local hot spots of high-power chips, characterized in that It includes the following steps: Obtain the heat source distribution and size specifications of the high-power chip. According to the heat source distribution and size specifications of the high-power chip, determine the heat source points, and set the initial spoiler element spacing according to the heat source power; Fit a parametric binary function in the XY plane based on the initial spoiler element spacing and the heat source points. Iterate the parametric binary function multiple times according to the average heat source temperature to determine the optimal spoiler element spacing; Determine the structure of the microchannel heat sink according to the optimal spoiler element spacing.

2. The method according to claim 1, wherein Iterating the parametric binary function multiple times according to the average heat source temperature to determine the optimal spoiler element spacing includes: Set the initial value of the control parameter in the parametric binary function and determine the iteration step size. Increase or decrease the control parameter to change the spacing function value, and recalculate the average heat source temperature. Determine whether the heat transfer capacity under the current parametric binary function is greater than that of the previous parametric binary function. When it is greater, continue to iterate until the heat transfer capacity less than that of the previous parametric binary function appears, determine the current control parameter, and continue to perform multiple rounds of iteration. If the heat transfer capacity under the current parametric binary function is greater than that of the previous round of parametric binary function, continue to iterate until the heat transfer capacity less than that of the previous parametric binary function appears, determine the current control parameter. If the situation that the heat transfer capacity under the current parametric binary function is greater than that of the previous round of parametric binary function does not appear until the maximum number of iterations is reached, output the current control parameter value and determine the optimal spoiler element spacing.

3. The method according to claim 2, wherein Calculate the average heat source temperature according to the heat transfer amount between the liquid and the microchannel heat sink under the microchannel heat sink structure. The heat transfer amount between the liquid and the microchannel heat sink under the microchannel heat sink structure is: Q = hAΔT where h is the convective heat transfer coefficient, A is the area of the interface between the liquid and the microchannel, and ΔT is a constant.

4. The method according to claim 1, wherein In determining the structure of the microchannel heat sink according to the optimal spoiler element spacing, the spoiler element is rectangular.

Images