Researchers in Japan have made a breakthrough in cooling technology by demonstrating how microfluidic channels integrated into AI processors and GPUs can effectively dissipate heat using a dual-phase approach. The team at the Institute of Industrial Science at the University of Tokyo implemented microchannels directly within the chip, enabling water to flow through and efficiently remove heat from the source through dual-phase cooling.
Traditionally, microfluidic cooling techniques have been limited by the heat capacity of water. By leveraging the latent heat of phase change of water, which is the thermal energy absorbed during boiling or evaporation, the researchers were able to achieve seven times the amount of cooling efficiency.
Lead researcher Hongyuan Shi explained, "By harnessing the latent heat of water, two-phase cooling becomes possible, resulting in a significant enhancement in heat dissipation efficiency." However, managing the flow of vapor bubbles as water transitions to steam presents a challenge, with factors such as microchannel geometry, two-phase flow regulation, and flow resistance playing crucial roles in maximizing heat transfer efficiency.
The innovative water-cooling system developed by the team features three-dimensional microfluidic channel structures with a capillary design and a manifold distribution layer. Various capillary geometries were designed and tested under different conditions to study their properties.
The manifolds in the system incorporate a two-stage flow design, where the coolant first enters spacious manifold channels, then passes through narrow microchannels in contact with the heated surface before exiting through another manifold channel. This design controls the distribution of coolant within the microchannel, optimizing heat dissipation.
Comparative analysis of flow boiling performance was conducted between a microchannel without a manifold and four samples with different manifold designs. The results showed that both the microchannel geometry and the manifold channels significantly impact the thermal and hydraulic performance of the system, with the coefficient of performance (COP) reaching up to 100,000, far surpassing conventional cooling techniques.
According to Masahiro Nomura from the Institute, "Effective thermal management of high-power electronic devices is essential for advancing next-generation technology, and our innovative design could pave the way for achieving the necessary cooling capabilities." The microchannels, which could be etched in silicon and wafer bonded to the back of chips, are specifically tailored to target hotspot areas in AI chips, offering enhanced cooling solutions for AI data centers.
The research paper detailing this groundbreaking cooling technology can be found at 10.1016/j.xcrp.2025.102520.