A heat dissipation stack structure of an AI computing card PCB

Abstract

The utility model relates to a kind of heat dissipation stack structures of AI computing power card PCB, including the power supply daughterboard and signal daughterboard of press bonding, the power supply daughterboard includes 2 layers first copper layer and at least 6 layers thick copper layer between the first copper layer, the thickness of the thick copper layer is 2oz, first dielectric layer of 3mil thickness is equipped between the first copper layer and thick copper layer and between the thick copper layer of adjacent two layers, the copper foil layer in the middle of the signal daughterboard the signal daughterboard includes 2 layers second copper layer and at least 10 layers thin copper layer between the second copper layer, the thickness of the thin copper layer is 0.5oz.The utility model has the beneficial effect in that high current carrying capacity can be guaranteed while optimizing the heat dissipation path and reducing the risk of board warping caused by asymmetric stack.

Description

Technical Field

[0001] This utility model relates to the field of PCB technology, specifically to a heat dissipation stack structure for an AI computing card PCB. Background Technology

[0002] As the performance of artificial intelligence (AI) computing cards continues to improve, their power consumption and heat generation also increase significantly. Traditional PCB stack-up designs are clearly insufficient to meet the heat dissipation requirements of high-power-density chips, especially in high-current, high-voltage applications. Problems such as uneven heat dissipation, localized overheating, and interlayer dielectric breakdown are becoming increasingly prominent. Currently, the current-carrying heat dissipation design layer is mainly concentrated in the middle layer of the board, which cannot meet the high current-carrying capacity and rapid heat dissipation requirements of AI chips, resulting in excessively high local temperature rise and affecting chip performance and reliability. Utility Model Content

[0003] The purpose of this invention is to provide a heat dissipation stack structure for an AI computing card PCB that can optimize the heat dissipation path and reduce the risk of board warping caused by stack asymmetry while ensuring high current carrying capacity.

[0004] A heat dissipation stacked structure for an AI computing card PCB includes a power sub-board and a signal sub-board bonded together. The power sub-board includes two first copper layers and at least six thick copper layers located between the first copper layers. The thickness of the thick copper layers is 2 oz. A first dielectric layer with a thickness of 3 mil is provided between the first copper layers and the thick copper layers, as well as between adjacent thick copper layers. The signal sub-board includes two second copper layers and at least ten thin copper layers located between the second copper layers. The thickness of the thin copper layers is 0.5 oz.

[0005] In the above scheme, the AI ​​computing card chip is mounted on the first copper layer of the top layer of the power sub-board. The layers between the two first copper layers are thick copper layers with a thickness of 2oz, so that the thick copper layers in the middle are all current-carrying layers, which can play a role in rapid heat dissipation. The first dielectric layer between the first copper layer and the thick copper layer, as well as between two adjacent thick copper layers, is 3mil thick. This can prevent the first dielectric layer from breaking down or partially discharging under high voltage due to the excessive copper teeth of the thick copper layer under high current. The thickness of the thin copper layer in the middle of the signal sub-board is 0.5oz. The overall number of layers of the power sub-board is at least 4 fewer than that of the signal sub-board. The thickness of the thick copper layer in the middle of the power sub-board is 2oz, which can reduce the thickness difference between the signal sub-board and the power sub-board, thereby helping to reduce the problem of board warping when the power sub-board and the signal sub-board are laminated.

[0006] Furthermore, both the power supply sub-board and the signal sub-board have symmetrical structures.

[0007] In the above scheme, both the power supply sub-board and the signal sub-board have symmetrical structures, so that the board warping problem will not occur when the power supply sub-board or the signal sub-board is pressed together, thereby reducing the board warping problem when the power supply sub-board and the signal sub-board are pressed together in the future.

[0008] Furthermore, a second dielectric layer is provided between the second copper layer and the thin copper layer, as well as between two adjacent thin copper layers. Both the first dielectric layer and the second dielectric layer are two layers of PP semi-cured sheets.

[0009] In the above solution, the PP prepreg has good adhesion. During the pressing process, the two layers of PP prepreg can form a strong bond between the first copper layer and the thick copper layer, between adjacent thick copper layers, between the second copper layer and the thin copper layer, and between adjacent thin copper layers. This allows the copper foils of each layer of the power supply sub-board and the signal sub-board to be tightly bonded together to form an integral structure.

[0010] Furthermore, the PP prepreg in the first medium layer has a high resin content, with the resin content of the first medium layer being 75% to 80%.

[0011] In the above scheme, the residual copper rate of the thick copper layer is relatively low, which means that the distribution of the thick copper layer may be somewhat uneven. The PP semi-cured sheet with a resin content of 75%~80% has good fluidity and filling properties. During the pressing process, the resin can fully fill the gaps between the thick copper layers, making the structure of the power supply sub-board more uniform and dense. This helps to reduce the differences in structure and performance between the power supply sub-board and the signal sub-board, making the performance of the entire AI computing card PCB more stable and consistent.

[0012] Furthermore, the resin content of the PP semi-cured sheet in the first dielectric layer is higher than that of the PP semi-cured sheet in the second dielectric layer.

[0013] In the above scheme, the signal sub-board is mainly used for high-speed signal transmission, which has extremely high requirements for signal integrity and stability. The second dielectric layer PP prepreg with a lower resin content can reduce dielectric loss and reduce signal attenuation and distortion during transmission. This is because when the resin content is relatively low, the dielectric constant and loss factor of the dielectric are relatively small, which is more conducive to lossless transmission of high-speed signals, thereby improving the signal transmission quality of the signal sub-board and ensuring that the AI ​​computing card can accurately and efficiently process and transmit data. Under the premise of meeting the different performance requirements of the power sub-board and the signal sub-board, the reasonable allocation of resin content can avoid the excessive use of high-cost high resin content materials. By using PP prepreg with a relatively low resin content in the signal sub-board, the material cost is reduced while ensuring performance.

[0014] Furthermore, the material of the thick copper layer is reverse copper foil, and its surface roughness Rz≤5μm.

[0015] In the above scheme, the surface roughness of copper foil has a significant impact on signal loss in high-speed signal transmission scenarios. When the surface roughness is large, the signal will generate more scattering and reflection on the copper foil surface, resulting in increased signal energy loss, which in turn affects the integrity and transmission quality of the signal. However, by selecting reverse copper foil with a surface roughness Rz≤5μm, its smooth surface can effectively reduce signal scattering and reflection, reduce signal loss during transmission, ensure that high-speed signals can be transmitted stably and accurately, and improve the signal processing capability and data transmission efficiency of the AI ​​computing card.

[0016] Furthermore, the thickness of both the first copper layer and the second copper layer is 0.51 oz.

[0017] In the above scheme, impedance matching is crucial for high-speed signal transmission. A copper layer thickness of 0.51oz helps to achieve a stable and predictable characteristic impedance, reducing signal reflection and distortion. When a signal is transmitted on a circuit board, if the impedance is mismatched, it will cause partial reflection of the signal energy, which may interfere with subsequent signals and affect the accuracy and integrity of data transmission.

[0018] Furthermore, the thickness difference between the power supply sub-board and the signal sub-board is between 0 and 0.1 mm.

[0019] In the above solution, PCBs undergo temperature changes during manufacturing and use. Different materials will generate thermal stress due to their different coefficients of thermal expansion. Power and signal sub-boards typically contain multiple materials, such as copper foil and insulating layers. When the thickness difference between the two sub-boards is small, their thermal expansion rates are more similar when heated. If the thickness difference is too large, the thermal expansion of the thicker sub-board may be significantly greater than that of the thinner sub-board, resulting in greater thermal stress at the joint. This stress can cause the circuit board to bend and deform, i.e., board warping. Controlling the thickness difference to within 0.1mm can effectively reduce the internal stress caused by inconsistent thermal expansion and reduce the possibility of board warping.

[0020] This invention discloses a heat dissipation stack structure for an AI computing card PCB, which optimizes the heat dissipation path and reduces the risk of board warping caused by stack asymmetry while ensuring high current carrying capacity. The AI ​​computing card chip is mounted on the first copper layer of the top layer of the power sub-board. The layers between the two first copper layers are thick copper layers with a thickness of 2oz, making the thick copper layers in the middle current-carrying layers, which can play a role in rapid heat dissipation. The first dielectric layer between the first copper layer and the thick copper layer, as well as between two adjacent thick copper layers, is 3mil thick. This can prevent the first dielectric layer from breaking down or partially discharging under high voltage due to the excessive copper teeth of the thick copper layer under high current. The thickness of the thin copper layer in the middle of the signal sub-board is 0.5oz. The power sub-board has at least 4 fewer layers than the signal sub-board. The 2oz thickness of the thick copper layer in the middle of the power sub-board can reduce the thickness difference between the signal sub-board and the power sub-board, thereby helping to reduce the problem of board warping when the power sub-board and the signal sub-board are laminated. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the heat dissipation stack structure of an AI computing card PCB according to one embodiment.

[0022] Explanation of reference numerals in the attached diagram: 1. Power supply sub-board; 11. First copper layer; 12. Thick copper layer; 13. First dielectric layer; 14. Substrate; 2. Signal sub-board; 21. Second copper layer; 22. Thin copper layer; 23. Second dielectric layer. Detailed Implementation

[0023] The following will describe in further detail the heat dissipation stack structure of an AI computing card PCB according to the present invention, with reference to specific embodiments and accompanying drawings.

[0024] like Figure 1 As shown in a preferred embodiment, the heat dissipation stack structure of the AI ​​computing card PCB of this utility model includes a power sub-board 1 and a signal sub-board 2 connected by pressing. The power sub-board 1 includes two first copper layers 11 and at least six thick copper layers 12 located between the first copper layers 11. The thickness of the thick copper layers 12 is 2oz. A first dielectric layer 13 with a thickness of 3mil is provided between the first copper layers 11 and the thick copper layers 12 and between two adjacent thick copper layers 12. The signal sub-board 2 includes two second copper layers 21 and at least ten thin copper layers 22 located between the second copper layers 21. The thickness of the thin copper layers 22 is 0.5oz.

[0025] In this embodiment, the AI ​​computing card chip is mounted on the first copper layer 11 on the top layer of the power subboard 1. The layers between the two first copper layers 11 are thick copper layers 12, with a thickness of 2oz. This makes the thick copper layers 12 in the middle a current-carrying layer, which can play a role in rapid heat dissipation. The first dielectric layer 13 between the first copper layer 11 and the thick copper layer 12, as well as between two adjacent thick copper layers 12, is 3mil thick. This can prevent the first dielectric layer 13 from breaking down or partially discharging under high voltage due to the excessive copper teeth of the thick copper layer 12 under high current. The thickness of the thin copper layer 22 in the middle of the signal subboard 2 is 0.5oz. The power subboard 1 has at least 4 fewer layers than the signal subboard 2. The thickness of the thick copper layer 12 in the middle of the power subboard 1 is 2oz, which can reduce the thickness difference between the signal subboard 2 and the power subboard 1, thereby helping to reduce the problem of board warping when the power subboard 1 and the signal subboard 2 are pressed together.

[0026] The power supply sub-board 1 and signal sub-board 2 form an "N+M" stacked structure. The N board and the M board can each undertake different functions. The N board is dedicated to heat dissipation and current flow design. The L2 to L7 layers are designed with 2oz thick copper foil. By utilizing the good conductivity and thermal conductivity of copper, power distribution and heat conduction are efficiently realized, providing a stable power supply to heat-generating components and removing heat in a timely manner. The M board can focus on other functions such as signal transmission and logic processing, such as arranging various signal lines and control circuits. This can avoid mutual interference between different functions and enable each functional module to achieve the best design effect.

[0027] Reference Figure 1 The power supply subboard 1 has eight copper foil layers: L1, L2, L3, L4, L5, L6, L7 and L8. L1 and L8 are the first copper layers 11, L2, L3, L4, L5, L6 and L7 are thick copper layers 12 with a thickness of 2oz, the space between L2 and L3, between L4 and L5 and between L6 and L7 is the substrate 14, and the space between L1 and L2, between L3 and L4, between L5 and L6 and between L7 and L8 is the first dielectric layer 13.

[0028] like Figure 1 As shown, in some embodiments, both the power subboard 1 and the signal subboard 2 have symmetrical structures. Since both the power subboard 1 and the signal subboard 2 have symmetrical structures, board warping will not occur when pressing either the power subboard 1 or the signal subboard 2 together, thereby reducing the risk of board warping when subsequently pressing the power subboard 1 and the signal subboard 2 together.

[0029] Reference Figure 1 The power supply sub-board 1 is symmetrically distributed on both sides with the plane between L4 and L5 as the axis of symmetry; the signal sub-board 2 is symmetrically distributed on both sides with the plane between L14 and L15 as the axis of symmetry.

[0030] like Figure 1As shown, in some embodiments, a second dielectric layer 23 is provided between the second copper layer 21 and the thin copper layer 22, as well as between two adjacent thin copper layers 22. The first dielectric layer 13 and the second dielectric layer 23 are both two layers of PP prepreg. PP prepreg has good adhesion. During the lamination process, the two layers of PP prepreg can form a strong bond between the first copper layer 11 and the thick copper layer 12, between adjacent thick copper layers 12, between the second copper layer 21 and the thin copper layer 22, and between adjacent thin copper layers 22. This allows the copper foils of the power supply subboard 1 and the signal subboard 2 to be tightly bonded together to form an integral structure.

[0031] like Figure 1 As shown, in some embodiments, the PP prepreg of the first dielectric layer 13 is a 1060 type PP with high resin content. The residual copper rate of the thick copper layer is relatively low, which means that the distribution of the thick copper layer may have some unevenness. The PP prepreg with a resin content of 75% to 80% has good flowability and filling properties. During the lamination process, the resin can fully fill the gaps between the thick copper layers, making the structure of the power supply sub-board more uniform and dense. This helps to reduce the differences in structure and performance between the power supply sub-board and the signal sub-board, making the performance of the entire AI computing card PCB more stable and consistent.

[0032] like Figure 1 As shown, in some embodiments, the resin content of the PP prepreg in the first dielectric layer 13 is higher than that in the PP prepreg in the second dielectric layer 23. The signal subboard 2 is mainly used for high-speed signal transmission, requiring extremely high signal integrity and stability. The lower resin content of the second dielectric layer 23 PP prepreg can reduce dielectric loss and decrease signal attenuation and distortion during transmission. This is because a relatively low resin content results in a lower dielectric constant and loss factor, which is more conducive to lossless transmission of high-speed signals, thereby improving the signal transmission quality of the signal subboard 2 and ensuring that the AI ​​computing card can accurately and efficiently process and transmit data. Under the premise of meeting the different performance requirements of the power subboard 1 and the signal subboard 2, a reasonable allocation of resin content can avoid the excessive use of high-cost, high-resin-content materials. By using a PP prepreg with a relatively low resin content in the signal subboard 2, material costs are reduced while maintaining performance.

[0033] In this embodiment, the first medium layer can be selected from 106-type prepreg and 1060-type prepreg with a resin content of 75% to 80%, and the second medium layer can be selected from 1060-type prepreg and 1067-type prepreg with a resin content of 70% to 72%.

[0034] like Figure 1As shown, in some embodiments, the material of the thick copper layer 12 is selected as reverse copper foil, with a surface roughness Rz≤5μm. In high-speed signal transmission scenarios, the surface roughness of the copper foil has a significant impact on signal loss. When the surface roughness is large, the signal transmission on the copper foil surface will generate more scattering and reflection, resulting in increased signal energy loss, which in turn affects the integrity and transmission quality of the signal. However, by selecting reverse copper foil with a surface roughness Rz≤5μm, its smooth surface can effectively reduce signal scattering and reflection, reduce signal loss during transmission, ensure that high-speed signals can be transmitted stably and accurately, and improve the signal processing capability and data transmission efficiency of the AI ​​computing card.

[0035] like Figure 1 As shown, in some embodiments, the thickness of both the two first copper layers 11 and the two second copper layers 21 is 0.51 oz. Impedance matching is crucial in high-speed signal transmission. A copper layer thickness of 0.51 oz helps achieve a stable and predictable characteristic impedance, reducing signal reflection and distortion. When a signal is transmitted on a circuit board, impedance mismatch can cause partial reflection of signal energy, which may interfere with subsequent signals and affect the accuracy and integrity of data transmission.

[0036] like Figure 1 As shown, in some embodiments, the thickness difference between the power supply sub-board 1 and the signal sub-board 2 is between 0 and 0.1 mm. During manufacturing and use, PCBs experience temperature changes. Different materials will generate thermal stress due to their different coefficients of thermal expansion. The power supply sub-board 1 and signal sub-board 2 typically contain multiple materials, such as copper foil and insulating layers. When the thickness difference between the two sub-boards is small, their thermal expansion rates are more similar when heated. If the thickness difference is too large, the thermal expansion of the thicker sub-board may be significantly greater than that of the thinner sub-board, resulting in greater thermal stress at the joint. This stress can cause the circuit board to bend and deform, i.e., board warping. Controlling the thickness difference between the power supply sub-board 1 and the signal sub-board 2 to within 0.1 mm can effectively reduce internal stress caused by inconsistent thermal expansion and reduce the possibility of board warping.

[0037] This invention discloses the working principle and process of a heat dissipation stack structure for an AI computing card PCB. Core components on the AI ​​computing card, such as GPU and CPU chips, generate a large amount of heat during operation. This heat is first transferred to the copper layer L1 in direct contact with it. Due to the high thermal conductivity of copper, the heat will quickly diffuse within the copper layer, reducing the local temperature. The heat is conducted from L1 to the entire L1 layer, and then through the first dielectric layer 13 to the thick copper layer 12 of L2. The thick copper layers 12 of L2-L7 further conduct the heat to other layers. Since the cross-sectional area of ​​the thick copper layer 12 is large, it can bear more heat transfer, thereby quickly dispersing the heat from the heat source to a larger area.

[0038] In the description of this utility model, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this utility model and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this utility model.

[0039] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this utility model, "a plurality of" means two or more, unless otherwise explicitly specified.

[0040] In this utility model, unless otherwise explicitly specified and limited, the terms "installation," "connection," "joining," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this utility model according to the specific circumstances.

[0041] Although the description of this utility model has been given in conjunction with the specific embodiments described above, it is obvious to those skilled in the art that many substitutions, modifications, and variations can be made based on the above description. Therefore, all such substitutions, modifications, and variations are included within the spirit and scope of the appended claims.

Claims

1. A heat dissipation stacked structure for an AI computing power card PCB, characterized in that, The device includes a power sub-board and a signal sub-board that are bonded together. The power sub-board includes two first copper layers and at least six thick copper layers located between the first copper layers. The thickness of the thick copper layers is 2 oz. A first dielectric layer with a thickness of 3 mil is provided between the first copper layers and the thick copper layers, as well as between two adjacent thick copper layers. The signal sub-board has a copper foil layer in the middle. The signal sub-board includes two second copper layers and at least ten thin copper layers located between the second copper layers. The thickness of the thin copper layers is 0.5 oz.

2. The heat dissipation stacked structure of the AI ​​computing card PCB according to claim 1, characterized in that, Both the power supply subboard and the signal subboard have symmetrical structures.

3. The heat dissipation stacked structure of the AI ​​computing card PCB according to claim 1, characterized in that, A second dielectric layer is provided between the second copper layer and the thin copper layer, as well as between two adjacent thin copper layers. Both the first dielectric layer and the second dielectric layer are two layers of PP semi-cured sheets.

4. The heat dissipation stacked structure of the AI ​​computing card PCB according to claim 1, characterized in that, The thick copper layer is made of reverse copper foil with a surface roughness Rz≤5μm.

5. The heat dissipation stacked structure of the AI ​​computing card PCB according to claim 1, characterized in that, The thickness of both the first copper layer and the second copper layer is 0.51 oz.

6. The heat dissipation stacked structure of the AI ​​computing card PCB according to claim 1, characterized in that, The thickness difference between the power supply sub-board and the signal sub-board is between 0 and 0.1 mm.

Images