AI chip cold plate type liquid cooling device based on diamond-copper-titanium porous structure and preparation method of AI chip cold plate type liquid cooling device
By using a diamond-copper-titanium porous composite material and powder metallurgy technology, the problem of mismatch between thermal conductivity and thermal expansion in AI chip heat dissipation devices has been solved, achieving efficient and reliable heat dissipation, which is suitable for high power density AI chips.
Patent Information
- Application Number
- CN202511038953.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-28
- Publication Date
- 2025-11-14
AI Technical Summary
Existing AI chip heat dissipation devices suffer from limited thermal conductivity, high interfacial thermal resistance, mismatch between structure and thermal expansion of silicon chips, and difficulty in scaling up manufacturing processes, making it difficult to meet the heat dissipation requirements of high power density chips.
A composite material with a porous structure of diamond-copper-titanium is used to construct a through-pore coolant channel, which is combined with a sealed shell to form a liquid cooling circulation channel. The material is prepared by powder metallurgy, achieving efficient bonding and structural optimization.
It significantly improves thermal conductivity, reduces interfacial thermal resistance, optimizes thermal expansion matching and structural strength, and enhances heat dissipation efficiency and reliability, making it suitable for efficient thermal management of high power density AI chips.
Smart Images

Figure CN120955055A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semiconductor heat dissipation technology, and in particular to a cold plate liquid cooling device suitable for high power density artificial intelligence (AI) chips, specifically a cold plate liquid cooling device for AI chips based on a diamond-copper-titanium porous structure and its preparation method. Background Technology
[0002] With the development of artificial intelligence (AI), big data, and high-performance computing, chip integration and power density continue to increase. The power consumption of a typical AI chip has exceeded 1000 W, and its heat flux density often exceeds 500 W / cm², placing extremely high demands on the heat dissipation system. Traditional air-cooling methods are no longer sufficient to meet the thermal management needs of high heat flux density chips. Their low heat dissipation efficiency and poor power utilization result in an overall system PUE (Power Usage Effectiveness) that is often higher than 1.5, which is detrimental to energy-efficient operation in high-performance computing scenarios such as data centers.
[0003] Liquid cooling with cold plates, as a relatively mature and efficient thermal management solution, has gradually replaced air cooling in some high-end servers and data centers. Existing cold plate structures are typically made of copper or aluminum, which, while possessing some thermal conductivity, have physical limitations. For example, copper's thermal conductivity is approximately 400 W / (m·K), and the interfacial thermal resistance between the metal and the chip is generally high, typically exceeding 2.5 mm²·K / W, making it difficult to meet the heat dissipation requirements of next-generation high-power chips. Furthermore, the mismatch in thermal expansion coefficients between the cold plate material and the silicon-based chip can also lead to thermal stress concentration, affecting long-term stable operation.
[0004] To improve the thermal conductivity of cold plates, some technical solutions have attempted to incorporate diamond materials. For example, Huawei's published patent CN119069436A proposes a structural design that incorporates diamond materials into the heat dissipation layer to enhance thermal conductivity. However, this solution mainly uses a stacked structure of diamond and passivation layers, which still suffers from problems such as insufficient bonding strength between diamond and the metal substrate, complex manufacturing process, high cost, and poor structural scalability. Therefore, a scalable engineering solution has not yet been developed.
[0005] Furthermore, Zhengzhou Shilu Technology proposed a diamond cooling system based on microwave plasma chemical vapor deposition (MPCVD) technology in patent CN222411821U for cooling semiconductor devices. Although this solution involves the thermal management application of diamond materials, it is mainly applied inside the deposition system and does not involve the design of the cold plate structure itself or the implementation of the metal composite structure, making it difficult to directly apply to the field of AI chip heat dissipation.
[0006] Other diamond heat sinks, such as those provided by Huanghe Whirlwind Company, have high thermal conductivity, but they usually adopt a dense diamond structure and do not form a porous gradient composite with the metal. They also do not achieve matching of the material's thermal expansion coefficient and interfacial metallurgical bonding. Therefore, they still cannot meet the comprehensive requirements for thermal conductivity, mechanical strength and manufacturing process under high heat flux density conditions.
[0007] Therefore, there is an urgent need for a cold plate-type liquid cooling device with high thermal conductivity, low interfacial thermal resistance, good structural scalability, matching with the thermal expansion performance of the chip, and controllable manufacturing process, in order to achieve efficient thermal management of high-power AI chips and improve the overall performance, stability and reliability of the heat dissipation system. Summary of the Invention
[0008] This invention aims to address the problems of limited thermal conductivity, high interfacial thermal resistance, mismatch between structure and thermal expansion of silicon chips, and difficulty in scaling up manufacturing processes in existing AI chip heat dissipation devices. It provides an AI chip cold plate-type liquid cooling device based on a diamond-copper-titanium porous structure and its fabrication method. By constructing a diamond-copper-titanium composite heat dissipation substrate with a gradient porous structure, combined with a sealed shell and liquid cooling circulation channels, this invention not only significantly improves thermal conductivity and reduces interfacial thermal resistance but also optimizes the overall thermal expansion matching and structural strength of the heat dissipation device, thereby achieving efficient and reliable thermal management of high-power-density AI chips.
[0009] The technical solution of this application is as follows: A liquid cooling device for AI chips based on a diamond-copper-titanium porous structure, characterized in that it includes: The heat dissipation substrate is a block made of a composite material composed of diamond particles, copper powder and titanium powder, and has through pores inside to form coolant channels. A casing is provided to seal and cover the heat dissipation substrate. Coolant inlet and outlet are respectively provided at both ends of the casing to form a closed liquid cooling circulation channel.
[0010] Preferably, the raw material composition of the heat dissipation substrate includes 40-60% diamond, 35-55% metal, and 5-10% auxiliary binder by mass percentage.
[0011] Preferably, the metal component consists of Cu powder and Ti powder, with each component having a mass percentage of 80-98% Cu powder and 2-20% Ti powder.
[0012] Preferably, the copper powder has a particle size of 10 μm to 100 μm, and the titanium powder has a particle size of 10 μm to 100 μm. Preferably, the volume composition of the heat dissipation substrate is: 20-45% diamond, 10-25% metal compound, and 30-70% porosity.
[0013] Preferably, the auxiliary adhesive is a mixture of acrylic emulsion and polyvinyl alcohol solution.
[0014] Preferably, the coolant is water or liquid metal.
[0015] A method for fabricating an AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure, characterized by the following steps: (1) Mix diamond particles, copper powder and titanium powder by mass percentage, with the volume fraction of diamond particles being 40%~60%, the volume fraction of copper powder being 30%~50%, and the volume fraction of titanium powder being 2%~10%, and add an appropriate amount of auxiliary binder and mix evenly; (2) Dry the mixture obtained in step (1) at 100~150℃ for 3~4 hours, and then obtain a uniform mixed powder by sieving; (3) The mixed powder is evenly added into the mold in batches and held under pressure of 40~120 MPa for 30~60 min by a hydraulic press; (4) Place the formed heat dissipation substrate blank under an argon protective atmosphere and sinter it. Heat it to 800-1050℃ at 3-10℃ / min, hold it for 30-120 min, and then cool it to room temperature in the furnace. (5) Trim the surface of the sintered heat dissipation substrate and remove burrs; (6) The heat dissipation substrate is embedded in a matching shell and fixed by welding or threaded connection, and the coolant inlet and outlet are sealed to form a liquid cooling circulation loop.
[0016] Preferably, the sintering heating rate is 5~10℃ / min, the sintering temperature is 950℃, and the holding time is 30 minutes.
[0017] Preferably, the casing is made of aluminum alloy or stainless steel, and sealing rings are provided at the connection holes of the coolant inlet and outlet to prevent coolant leakage.
[0018] Technical effect Compared with existing technologies, the AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure of the present invention has the following significant technical effects: Achieving a highly efficient heat exchange structure design: This invention introduces a through-pore structure into a diamond-copper-titanium composite material to construct a coolant channel through which the coolant can flow directly, thereby realizing direct heat exchange between the liquid inside the heat dissipation substrate and the heat source. This significantly improves the efficiency of convection and boiling heat transfer, and is significantly superior to the traditional surface heat exchange type cold plate structure.
[0019] Significantly improves thermal conductivity and heat dissipation efficiency: Using highly thermally conductive diamond particles as a thermally conductive framework, combined with copper powder to construct a metal thermally conductive network, and titanium powder to strengthen the interfacial metallurgical bond, the overall thermal conductivity of the composite material can reach 1200-1500 W / (m·K) with a porosity of 30%-70%. The chip junction temperature can be reduced by more than 20℃, effectively coping with high heat load scenarios with a single chip heat flux density ≥500 W / cm².
[0020] Matching thermal expansion properties and reducing interfacial thermal resistance: The porous structure and the introduction of titanium powder enable the thermal expansion coefficient of the composite material to be controlled at 6~8 ppm / K, which can be well matched with silicon-based chips and avoid thermal stress concentration. At the same time, the thermal resistance of the metallurgical bonding interface can be lower than 0.5 mm²·K / W, which is far superior to the existing copper / aluminum cold plate solution.
[0021] The process parameters are clear, controllable, and suitable for mass production. This invention achieves structural forming through conventional powder metallurgy processes, including mixing, cold pressing, sieving, sintering under argon protection, and post-processing. The entire process is completed within the range of 100~1050℃, making it suitable for industrial manufacturing needs.
[0022] Balancing strength and corrosion resistance, with high reliability: The heat dissipation substrate has a pressure resistance of ≥80 MPa, which can withstand the operating pressure of the liquid cooling system; the material itself is resistant to chloride ion corrosion concentration >500 ppm, and is suitable for long-term stable operation in high humidity, high heat and high load environments in data centers.
[0023] Material and structural cost optimization design: By controlling the diamond in the key heat conduction path area and using metal components in other areas, the material is utilized in a gradient manner, effectively reducing the overall material cost to about 20% of the all-diamond solution, and has excellent cost performance.
[0024] The system has strong integration capabilities and is compatible with a variety of coolants. The device of this invention can be adapted to water or liquid metal coolant, supports operating conditions of 60~80℃, and is suitable for various scenarios such as AI servers, edge computing, and high-performance computing clusters. The system PUE can be controlled below 1.05.
[0025] In this technical solution, during the sintering process, titanium reacts with the surface of diamond particles to form a dense TiC layer, reducing the interfacial thermal resistance to below 0.5 mm²·K / W. Simultaneously, the through-pore liquid cooling channels significantly increase the contact area between the coolant and the thermally conductive framework, achieving highly efficient two-phase heat exchange. Test results show that the overall thermal conductivity of the cold plate of this invention reaches 1200–1500 W / (m·K), and the chip junction temperature is reduced by more than 20°C in typical AI chip full-load tests. Attached Figure Description
[0026] Figure 1 This is a schematic diagram of the structure of the AI chip cold plate liquid cooling device based on the diamond-copper-titanium porous structure of this technical solution; Figure 2 This is a breakdown diagram of the technical solution.
[0027] 1-Top cover; 2-Heat dissipation base; 3-Covering shell; 4-Coolant inlet. Detailed Implementation
[0028] This cold-plate liquid cooling device mainly consists of an upper cover plate 1, a heat dissipation substrate 2, a protective shell 3, and coolant inlets 4. The upper cover plate 1 seals and secures the structure above the device, ensuring no coolant leakage. The heat dissipation substrate 2 is made of a diamond-copper-titanium porous composite material with internal through-pores for heat conduction and as a flow channel for the coolant, enabling efficient heat exchange between the chip and the coolant. The protective shell 3 covers and supports the heat dissipation substrate, forming an external enclosed space for the liquid cooling channel and providing structural strength. The coolant inlets 4 are located on both sides of the shell for connecting the coolant inlet and outlet, enabling liquid cooling circulation within the cold plate.
[0029] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the present invention is not limited to the following embodiments.
[0030] Example 1: Preparation of a typical cold plate under moderate heat flux density This embodiment is applicable to the heat dissipation needs of conventional AI inference chips, GPU modules, or edge devices.
[0031] Raw material ratio and particle size Diamond particles: 50% by mass, particle size 40–60 μm; Copper powder: 45% by mass, particle size 50–80 μm; Titanium powder: 5% by mass, particle size 40–60 μm; Auxiliary adhesive: accounting for approximately 7% by mass, composed of acrylic emulsion and polyvinyl alcohol solution in a volume ratio of 1:1.
[0032] Forming and sintering Cold pressing pressure: 80 MPa, holding pressure for 30 min; Drying temperature: 120℃, 3 hours; Sintering conditions: Under an argon protective atmosphere, the heating rate is 5℃ / min, the final temperature is 950℃, the holding time is 30 minutes, and the furnace is cooled to room temperature.
[0033] Porosity and structural parameters The porosity is 45%; The average pore size is approximately 100 μm; Heat sink base dimensions: 50 mm × 50 mm × 3 mm; Coolant: Deionized water; flow rate 1 L / min, pressure drop <15 kPa.
[0034] Packaging method The housing is made of aluminum alloy and uses threaded connections with sealing rings.
[0035] The device was successfully used in a mid-range AI inference accelerator, running continuously for 72 hours without leakage or abnormal temperature rise.
[0036] Example 2: High heat flux density cold plate device (enhanced type) Suitable for cooling requirements of high-end AI training chips (such as H100) or massively parallel computing modules.
[0037] Raw material ratio and particle size Diamond: 60% by mass, particle size 20–40 μm; Copper powder: 38%, particle size 10–40 μm; Titanium powder: 2%, particle size 10–20 μm; The auxiliary binder is high-concentration polyvinyl alcohol, accounting for 5% by mass.
[0038] process parameters Cold pressing: 120 MPa, holding pressure for 60 minutes; Drying temperature: 150℃, 4 hours; Sintering heating rate: 10℃ / min; sintering temperature: 1050℃, holding time: 60 minutes.
[0039] Performance and Structure With a porosity of 30%, the channels are more compact; The coolant is liquid gallium-based metal (non-conductive), with a flow rate of 1.8 L / min; The housing is made of stainless steel, laser-welded, and the connection port is equipped with a sealing ring.
[0040] In high-load continuous computing scenarios, the chip junction temperature is controlled below 70°C, which is significantly better than traditional copper water cooling plates.
[0041] Example 3: Low heat flux density economical cold plate (small edge application) This embodiment addresses the cooling needs of low-cost AI modules, consumer products, and small edge devices.
[0042] Raw material ratio Diamond 40%, copper powder 55%, titanium powder 5%, binder 7%.
[0043] The particle size of each raw material is controlled between 80 and 100 μm, resulting in lower manufacturing costs.
[0044] Process conditions Drying temperature: 100℃, 3 hours; Cold pressing pressure: 40 MPa, holding pressure for 30 minutes; The sintering temperature is 800℃, the heating rate is 3℃ / min, and the holding time is 60 minutes.
[0045] Structural features The heat dissipation substrate is relatively thin, approximately 2 mm; With a porosity of approximately 60%, heat dissipation is primarily aided by natural convection. The housing is made of injection-molded aluminum-plastic composite material, making it suitable for cost-sensitive equipment.
[0046] This solution is used in smart cameras and edge recognition devices, providing sufficient heat dissipation while maintaining good manufacturability.
Claims
1. A liquid cooling device for AI chips based on a diamond-copper-titanium porous structure, characterized in that: include: The heat dissipation substrate is a block made of a composite material composed of diamond particles, copper powder and titanium powder, and has through pores inside to form coolant channels. A casing is provided to seal and cover the heat dissipation substrate. Coolant inlet and outlet are respectively provided at both ends of the casing to form a closed liquid cooling circulation channel.
2. The AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure according to claim 1, characterized in that: The raw material composition of the heat dissipation substrate includes, by mass percentage, 40-60% diamond, 35-55% metal, and 5-10% auxiliary binder.
3. The AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure according to claim 1, characterized in that: The metal composition consists of Cu powder and Ti powder, with the following mass percentages: Cu powder 80-98% and Ti powder 2-20%.
4. The AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure according to claim 1, characterized in that: in, The copper powder has a particle size of 10 μm to 100 μm, and the titanium powder has a particle size of 10 μm to 100 μm.
5. The AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure according to claim 1, characterized in that: The volume composition of the heat dissipation substrate is: 20-45% diamond, 10-25% metal compound, and 30-70% porosity.
6. The AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure according to claim 1, characterized in that: The auxiliary binder is a mixture of acrylic emulsion and polyvinyl alcohol solution.
7. The AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure according to claim 1, characterized in that: in, The coolant is water or liquid metal.
8. A method for fabricating an AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure, characterized in that: Includes the following steps: (1) Mix diamond particles, copper powder and titanium powder by mass percentage, with the volume fraction of diamond particles being 40%~60%, the volume fraction of copper powder being 30%~50%, and the volume fraction of titanium powder being 2%~10%, and add an appropriate amount of auxiliary binder and mix evenly; (2) Dry the mixture obtained in step (1) at 100~150℃ for 3~4 hours, and then obtain a uniform mixed powder by sieving; (3) The mixed powder is evenly added into the mold in batches and held under pressure of 40~120 MPa for 30~60 min by a hydraulic press; (4) Place the formed heat dissipation substrate blank under an argon protective atmosphere and sinter it. Heat it to 800-1050℃ at 3-10℃ / min, hold it for 30-120 min, and then cool it to room temperature in the furnace. (5) Trim the surface of the sintered heat dissipation substrate and remove burrs; (6) The heat dissipation substrate is embedded in a matching shell and fixed by welding or threaded connection, and the coolant inlet and outlet are sealed to form a liquid cooling circulation loop.
9. The method for fabricating an AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure according to claim 8, characterized in that: in, The sintering heating rate is 5~10℃ / min, the sintering temperature is 950℃, and the holding time is 30 minutes.
10. The method for fabricating an AI chip cold plate liquid cooling device based on a diamond-copper-titanium porous structure according to claim 8, characterized in that: in, The casing is made of aluminum alloy or stainless steel, and sealing rings are provided at the connection holes of the coolant inlet and outlet to prevent coolant leakage.
Citation Information
Patent Citations
High heat-conductive diamond-copper composite encapsulating material and method for making same
CN101139515A
Radar antenna array surface non-uniform heat flux heat dissipation structure based on liquid metal
CN119852675A
Heat-dissipating structure and method for fabricating the same
US20110083835A1
High Efficiency Thermal Management Device for Use With Components Having High Heat Flux Values
US20200126892A1