A chip-level cooling and direct air capture integrated system and method based on carbon dioxide phase change cycle

Abstract

The application discloses a chip-level cooling and direct air capture integrated system and method based on carbon dioxide phase change cycle, and belongs to the technical field of data center thermal management and carbon capture. The system comprises a chip-level CO2 cooling module, a gaseous CO2 waste heat driven DAC module, a CO2 liquefaction regeneration module and an intelligent control module. After CO2 absorbs the high heat flux heat of an AI chip in a micro-channel cold plate and is phase changed into high-temperature gas, the high-temperature gas directly enters a DAC desorption tower as a regeneration heat source to drive a solid adsorbent to desorb, and the captured high-concentration CO2 is compressed and liquefied to be injected into the cooling module, thereby forming a "cooling-capture-regeneration" closed loop. The application utilizes the excellent heat transfer characteristics of CO2 in the pseudo-critical region to realize chip-level efficient cooling, and the PUE can be as low as 1.02 or below; meanwhile, the phase change waste heat is directly used to drive the DAC, and the single-ton CO2 capture energy consumption is reduced to 3.2 GJ, which is reduced by 40-50% compared with the traditional scheme. The system has an annual CO2 capture capacity of 50-100 thousand tons, realizes the transition of a data center from "zero carbon" to "negative carbon", and is suitable for AI training centers, supercomputing centers and zero-carbon parks with a capacity of 100 MW or above.

Description

Technical Field

[0001] This invention belongs to the field of data center thermal management and carbon capture technology, specifically involving a closed-loop system that deeply integrates AI chip-level high-efficiency cooling with direct air capture (DAC), which is particularly suitable for zero-carbon / negative-carbon operation of AI computing centers with high heat flux density. Background Technology

[0002] With the explosive growth of generative AI, the thermal design power (TDP) of AI chips (such as the NVIDIA GB200) has exceeded 1200W, and the heat flux density has exceeded 500W / cm², pushing traditional air cooling and cold plate liquid cooling to their limits. While immersion liquid cooling (fluorinated liquid) can achieve a PUE of 1.03, it faces challenges such as high cost (fluorinated liquid accounts for 15% of OPEX), high global warming potential (GWP), and non-renewable nature.

[0003] Meanwhile, carbon emissions from data centers are becoming increasingly serious. Direct air capture (DAC) technology can remove CO2 from the atmosphere, but current DACs have extremely high energy consumption (6-10 GJ / t CO2), with adsorbent regeneration heat accounting for >60%, and typically require high-temperature steam of 90-120℃, resulting in DAC costs as high as $600-1000 / ton, which seriously restricts commercialization.

[0004] In existing technologies, CO2 cooling (such as transcritical CO2 heat pumps) and DACs are mostly independent systems. While companies like Microsoft and Meta are exploring using waste heat from data centers to drive DACs, they all employ traditional water / fluorinated liquid cooling. The waste heat (60-80°C) needs to be boosted by a heat pump before it can be used for DAC regeneration, resulting in complex systems with low energy efficiency. There is currently no integrated solution that directly uses CO2 as a chip cooling medium and utilizes its phase change heat to directly drive the DAC. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the dual challenges of high energy consumption and large carbon emissions of existing liquid cooling technology and high cost of DAC technology, and to provide a closed-loop system that deeply integrates AI chip-level high-efficiency cooling with DAC, so as to realize a negative carbon computing infrastructure that "captures carbon with carbon and generates heat with heat".

[0006] The technical solution of this invention is: System Architecture

[0007] A chip-level cooling and direct air capture integrated system and method based on carbon dioxide phase change cycle includes: 1. Chip-level CO2 cooling module (100)

[0008] Level CO2 cooling module (100) The device is mounted on the surface of an AI computing chip (200) and includes a microchannel cold plate (110) and a phase change cavity (120). Multiple microscale channels (111) are formed within the microchannel cold plate (110) using microfabrication techniques (such as photolithography-etching or 3D printing), with hydraulic diameters of 50-500 μm, preferably 100-200 μm.

[0009] The CO2 working fluid flows within the microscale channel (111) at an operating pressure of 7.5-15 MPa (higher than the critical pressure of 7.38 MPa) and an operating temperature of -40°C to 60°C. When CO2 flows through the chip hotspot, it absorbs heat and its temperature rises to around 31°C (pseudo-critical temperature). At this point, the specific heat capacity and thermal conductivity increase sharply, forming a "thermal overshoot" effect. The heat transfer coefficient can reach 3-5 times that of water, achieving efficient and uniform temperature cooling.

[0010] The microchannel cold plate (110) adopts a multi-layer stacked structure, with each layer of channels arranged in a serpentine or forked pattern. The layers are connected by micropores (112) to form a three-dimensional mesh channel structure. This design enables CO2 to form a uniform temperature distribution on the chip surface, avoiding local hot spots, and reducing the temperature non-uniformity index to 0.24 (compared to 4.22 for traditional water cooling). 2. Waste heat from gaseous CO2 drives the DAC module (300).

[0011] The gaseous CO2 pipeline (130) is connected to the outlet of the chip-level CO2 cooling module (100), and includes an adsorption tower (310), a desorption tower (320), and a solid adsorbent (330).

[0012] The key innovation lies in the direct use of gaseous CO2 as a regeneration heat source. The temperature of the CO2 flowing out from the chip outlet is 80-150℃ (depending on the chip load), which precisely matches the desorption and regeneration temperature (80-120℃) of the solid adsorbent (330). The gaseous CO2 directly enters the regeneration heat exchanger (321) of the desorption tower (320), where it exchanges heat with the CO2-loaded adsorbent, driving the desorption reaction. Adsorbent - CO2 + heat → Adsorbent + CO2↑

[0013] The solid adsorbent (330) is preferably an amino-functionalized mesoporous silica material (such as PEI / SBA-15) or a MOF material (such as Mg-MOF-74), with a CO2 adsorption capacity >2 mmol / g, a desorption temperature of 80-100℃, and a high degree of matching with the temperature grade of gaseous CO2, and a heat grade matching degree >85%.

[0014] The desorption tower (320) employs a variable temperature and pressure coupled regeneration process. Gaseous CO2 first enters the first-stage regeneration heat exchanger (321a) to provide sensible heat, preheating the adsorbent to 80°C; then it enters the second-stage regeneration heat exchanger (321b) to provide latent heat, maintaining the desorption temperature at 100-120°C. This cascaded utilization reduces the waste heat consumption per ton of CO2 captured to below 3.5 GJ. 3. CO2 liquefaction and regeneration module (400)

[0015] It includes a compressor (410), a condenser (420), and a storage tank (430). High-concentration CO2 (>90%) released from the desorption tower (320) is compressed to 8-10 MPa by the compressor (410), liquefied in the condenser (420) (critical temperature 31.1℃, liquefaction can occur at room temperature), and stored in the storage tank (430).

[0016] The storage tank (430) is connected to the inlet of the chip-level CO2 cooling module (100) via a liquid CO2 pipeline (140), forming a closed-loop circulation. The CO2 charge of the system is determined according to the scale of the data center; a 100MW data center requires approximately 50-100 tons of CO2 working fluid.

[0017] Preferably, the CO2 liquefaction and regeneration module (400) further includes an expander (440). After compression, the high-concentration CO2 first passes through the expander (440) to recover expansion work before entering the condenser (420). The expansion work can be used to drive the compressor (410) or for auxiliary cooling, improving the overall system efficiency by 5-8%. 4. Intelligent control module (500)

[0018] It includes a pressure sensor (510), a temperature sensor (520), a controller (530), and an AI prediction unit (540). The controller (530) dynamically adjusts the CO2 flow rate (via a variable frequency pump or valve), pressure (via compressor speed), and DAC desorption rhythm (via an adsorption tower switching valve) according to the chip's thermal load.

[0019] The AI ​​prediction unit (540) predicts the sudden load of AI training tasks (such as the gradient synchronization stage in large model training) 30-60 seconds in advance based on load prediction algorithms (such as LSTM neural networks) and adjusts the CO2 circulation flow in advance to avoid chip temperature overshoot. Method and Flow

[0020] A chip-level cooling and direct air capture integrated system and method based on carbon dioxide phase change cycle includes: S1: Liquid or supercritical CO2 enters the chip-level microchannel cold plate (110) at 7.5-15MPa and -40℃ to 35℃, absorbing the high heat flux density (>500W / cm²) of the AI ​​computing chip (200), and the phase changes to gaseous CO2 at 80-150℃; S2: The gaseous CO2 directly enters the regeneration heat exchanger (321) of the DAC desorption tower (320) as the sole heat source to drive the desorption and regeneration of the solid adsorbent (330). The high concentration CO2 (>90% concentration) released by the solid adsorbent (330) is collected. S3: The high-concentration CO2, after being compressed and condensed into liquefaction, is reinjected into the chip-level CO2 cooling module (100) to complete the closed-loop cycle; S4: Repeat steps S1-S3 to form a continuous "cooling-collection-regeneration" integrated cycle.

[0021] Preferably, in step S2, after the gaseous CO2 drives the DAC to regenerate, the temperature drops to 60-80°C, and then the adsorbent to be regenerated enters the adsorption tower (310) of the DAC module (300) after being preheated by the heat recovery heat exchanger (322), thereby realizing cascaded heat utilization and further improving energy efficiency.

[0022] Preferably, in step S3, the compression process employs multi-stage compression with interstage cooling, and the interstage cooling heat is used to preheat the adsorbent entering the desorption tower (320), reducing the overall energy consumption of the system by 10-15%.

[0023] Preferably, the method further includes step S5: when the AI ​​chip is under low load (such as inference tasks or at night), excess gaseous CO2 is introduced into the CO2 energy storage unit (700) to drive the turbine to generate electricity or store cold energy, thereby achieving peak-valley regulation. At this time, the system switches from the "cooling-capture" mode to the "cooling-storage" mode. Beneficial effects

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Breakthrough in energy efficiency

[0025] This invention utilizes the latent heat of CO2 phase change (574 kJ / kg) to directly drive DAC regeneration, eliminating the need for heat pumps or steam boilers in traditional solutions, thus reducing DAC energy consumption by 40-50%. Simultaneously, chip cooling employs supercritical CO2 microchannel technology, achieving a PUE as low as 1.02, a further improvement over traditional fluorinated liquid immersion cooling (PUE 1.03-1.09). 2. Carbon negative capacity

[0026] A 100MW data center consumes 876 GWh of electricity annually and generates approximately 350 GWh of waste heat. This system can utilize 60-70% of this waste heat to drive the DAC (Data Center Automation), capturing 50,000-100,000 tons of CO2 annually. This not only achieves net-zero emissions for the data center but also generates "negative carbon computing power," producing carbon credit revenue (based on 100 yuan / ton, with annual revenue of 5-10 million yuan). 3. Water-Energy Synergy

[0027] While capturing CO2 from the atmosphere, the DAC process also condenses a large amount of water (0.5-1 ton of water per ton of CO2). This system is equipped with a water recovery unit (600) that utilizes the residual cooling of gaseous CO2 to condense atmospheric moisture and generate fresh water to supply cooling water to the data center, saving 200,000 tons of water annually and solving the water consumption problem of the data center. 4. Material compatibility

[0028] CO2 is chemically inert and does not corrode chip packaging materials or PCB substrates. Its long-term reliability is superior to water cooling (which carries the risk of leakage and short circuit) and fluorinated liquids (which carry the risk of material swelling). The high-pressure seal uses metal-ceramic composite technology, referencing aerospace-grade standards, to ensure long-term stable operation at 8-10 MPa. 5. Modular design and rapid deployment

[0029] The system adopts a modular design, with each module supporting 10-50kW racks and enabling rapid deployment via a containerized configuration. It is compatible with existing data center infrastructure, supports hot-swappable maintenance, and does not affect the continuity of AI training tasks. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention;

[0031] Figure 2 This is a schematic diagram of the microchannel cold plate structure for a chip-level CO2 cooling module.

[0032] Figure 3 This is a schematic diagram illustrating the working principle of a DAC module driven by waste heat from gaseous CO2.

[0033] Figure 4 A flowchart of the CO2 liquefaction and regeneration module;

[0034] Figure 5 This is the control logic diagram for the intelligent control module;

[0035] Figure 6 This is a diagram illustrating the switching of system operating modes. Detailed Implementation Example 1: 100MW AI Training Data Center

[0036] A 100MW AI training data center is equipped with 10,000 NVIDIA GB200 servers, with a single rack power density of 50kW. The system of this invention is used in this application. Chip-level CO2 cooling module: Each server is equipped with a microchannel cold plate with a hydraulic diameter of 150μm, a CO2 flow rate of 0.5kg / s, an inlet pressure of 10MPa, a temperature of 25℃, and an outlet temperature of 120℃; DAC module: Equipped with 10 sets of parallel adsorption towers, each set with an adsorbent loading of 10 tons, using PEI / SBA-15 adsorbent, CO2 capture capacity of 2.5 mmol / g, desorption temperature of 100℃, and single tower capture capacity of 10,000 tons of CO2 / year. Liquefaction module: Employs a two-stage compression + expander recovery system. The compressor consumes 15kW, the expander recovery system consumes 5kW, resulting in a net power consumption of 10kW and a liquefaction efficiency of >95%. Control module: The AI ​​prediction unit is based on the TensorFlow framework, with a prediction accuracy of >90% and a temperature control accuracy of ±1℃.

[0037] Execution result: PUE: 1.015 (cooling system energy consumption accounts for only 1.5% of IT load); Annual CO2 capture: 80,000 tons (utilizing 70% waste heat); Annual water savings: 150,000 tons; Carbon credit income: RMB 8 million / year (based on RMB 100 / ton). Example 2: Overclocking Scenario in a Supercomputing Center

[0038] A national supercomputing center needs to overclock its GPUs, achieving a heat flux density of 800 W / cm². The supercritical CO2 mode of this invention's system is employed. Operating pressure 12MPa, CO2 temperature 35℃ (supercritical region), operating in the peak specific heat capacity region; The microchannel has a hydraulic diameter of 100 μm and a heat transfer coefficient of 50,000 W / (m²·K). Chip surface temperature uniformity is ±0.5℃, and overclocking stability is improved by 30%.

Claims

1. A chip-level cooling and direct air capture integrated system and method based on carbon dioxide phase change cycle, characterized in that, include: A chip-level CO2 cooling module (100) is configured on the surface of an AI computing chip (200) and includes a microchannel cold plate (110) and a phase change cavity (120). Multiple microscale channels (111) are formed in the microchannel cold plate (110). After absorbing heat from the chip in the microscale channels (111), the CO2 changes from a liquid or supercritical state to a gaseous state. The gaseous CO2 waste heat drives the DAC module (300), which is connected to the outlet of the chip-level CO2 cooling module (100) through the gaseous CO2 pipeline (130). It includes an adsorption tower (310), a desorption tower (320) and a solid adsorbent (330). The gaseous CO2 is used as a regeneration heat source and directly enters the regeneration heat exchanger (321) of the desorption tower (320) to drive the solid adsorbent (330) to desorb, regenerate and release high-concentration CO2. The CO2 liquefaction and regeneration module (400) includes a compressor (410), a condenser (420), and a storage tank (430). The high-concentration CO2 is compressed by the compressor (410), liquefied in the condenser (420), and stored in the storage tank (430). The storage tank (430) is connected to the inlet of the chip-level CO2 cooling module (100) through a liquid CO2 pipeline (140) to form a closed-loop circulation. The intelligent control module (500) includes a pressure sensor (510), a temperature sensor (520), and a controller (530). The controller (530) dynamically adjusts the CO2 flow rate, pressure, and DAC desorption rhythm according to the chip's thermal load to maintain the steady-state operation of the system.

2. The system according to claim 1, characterized in that, The hydraulic diameter of the microscale channel (111) is 50-500 μm. The operating pressure of CO2 in the microscale channel (111) is 7.5-15 MPa, and the operating temperature is -40℃ to 60℃. The peak specific heat capacity of CO2 in the pseudocritical region (around 31℃) is used to enhance heat transfer.

3. The system according to claim 1 or 2, characterized in that, The microchannel cold plate (110) adopts a multi-layer stacked structure. Each layer of the microscale flow channel (111) is arranged in a serpentine or cross-flow pattern. The layers are connected by micropores (112) to form a three-dimensional mesh flow channel structure to homogenize the temperature distribution on the chip surface.

4. The system according to claim 1, characterized in that, The solid adsorbent (330) is an amino-functionalized mesoporous material or a metal-organic framework (MOF) material, and its desorption and regeneration temperature is 80-120°C. The temperature of the gaseous CO2 at the outlet of the chip-level CO2 cooling module (100) is 80-150°C, and the heat quality matching degree is >85%.

5. The system according to claim 1, characterized in that, The desorption tower (320) adopts a variable temperature and pressure coupled regeneration process. The gaseous CO2 first enters the first-stage regeneration heat exchanger (321a) to provide sensible heat, and then enters the second-stage regeneration heat exchanger (321b) to provide latent heat, so as to realize the cascade utilization.

6. The system according to claim 1, characterized in that, The CO2 liquefaction and regeneration module (400) also includes an expander (440). After the high-concentration CO2 is compressed by the compressor (410), it first passes through the expander (440) to recover the expansion work, and then enters the condenser (420). The expansion work is used to drive the compressor (410) or for auxiliary cooling.

7. The system according to claim 1, characterized in that, The system also includes a water recycling unit (600), which is located at the exhaust end of the DAC module (300). It utilizes the residual heat after the gaseous CO2 is cooled to condense atmospheric moisture and generate fresh water to supply cooling water to the data center.

8. The system according to claim 1, characterized in that, The intelligent control module (500) also includes an AI prediction unit (540), which adjusts the CO2 circulation flow rate 30-60 seconds in advance based on a load prediction algorithm to cope with the sudden load of the AI ​​training task.

9. A chip-level cooling and direct air capture integrated system and method based on carbon dioxide phase change cycle, characterized in that, Includes the following steps: S1: Liquid or supercritical CO2 enters the chip-level microchannel cold plate (110) at 7.5-15MPa and -40℃ to 35℃, absorbing the high heat flux density (>500W / cm²) of the AI ​​computing chip (200), and the phase changes to gaseous CO2 at 80-150℃; S2: The gaseous CO2 directly enters the regeneration heat exchanger (321) of the DAC desorption tower (320) as the sole heat source to drive the desorption and regeneration of the solid adsorbent (330). The high concentration CO2 (>90% concentration) released by the solid adsorbent (330) is collected. S3: The high-concentration CO2, after being compressed and condensed into liquefaction, is reinjected into the chip-level CO2 cooling module (100) to complete the closed-loop cycle; S4: Repeat steps S1-S3 to form a continuous "cooling-collection-regeneration" integrated cycle.

10. The method according to claim 9, characterized in that, In step S2, after the gaseous CO2 drives the DAC to regenerate, the temperature drops to 60-80℃, and then the adsorbent to be regenerated enters the adsorption tower (310) of the DAC module (300) after being preheated by the heat recovery heat exchanger (322), thereby realizing cascaded heat utilization.

11. The method according to claim 9, characterized in that, In step S3, the compression process employs multi-stage compression with interstage cooling. The interstage cooling heat is used to preheat the adsorbent entering the desorption tower (320), thereby reducing the overall energy consumption of the system.

12. The method according to claim 9, characterized in that, The method further includes step S5: when the AI ​​chip is under low load, the excess gaseous CO2 is introduced into the CO2 energy storage unit (700) to drive the turbine to generate electricity or store cold energy, thereby achieving peak-valley regulation.

Images