In the summer of 2026, we were asked to develop a high-performance single-phase liquid-cooling solution for a high-density chip with an average heat flux of approximately 300 W/cm² and local hotspots reaching 608 W/cm². The engineering boundaries were explicit: 40°C coolant at the inlet, a maximum junction temperature no higher than 105°C, and the lowest practical pressure drop.
From confirmed requirements through design, manufacturing, and test validation, the complete delivery took 18 days.
The result: with 40°C inlet water, the cold plate achieved an overall pressure drop of approximately 7 kPa while keeping the maximum chip junction temperature within 105°C. Every target was met using single-phase liquid cooling.
1. The real question was not simply whether it could be cooled
At a local heat flux of 608 W/cm², making the cold plate larger or simply increasing flow is not a viable answer. More flow can intensify local heat transfer, but it also raises pressure drop, pump power, and the burden on the entire cooling system.
We separated the challenge into two tightly coupled fields: how quickly the thermal field could spread heat, and how precisely the flow field could cover hotspots. We call the method dual-field precision engineering. Its purpose is not to maximize heat transfer and flow independently, but to balance local cooling capacity, pressure drop, flow distribution, and manufacturability within real project boundaries.
For AI servers and high-performance computing, cooling above 150 W/cm² has long depended on capabilities concentrated among overseas suppliers. Coupled multiphysics simulation, micro-scale manufacturing, and advanced composite materials each require sustained engineering accumulation. This project was designed to cross that barrier directly.
2. Thermal field: put cooling capacity exactly where the heat is
The difficulty of cooling a high-heat-flux chip is governed less by average power than by the location, area, and intensity of its hotspots.
We reconstructed the thermal model from the actual source map and designed a distributed jet-impingement array around the hotspots. Coolant is not spread uniformly; high-speed jets are directed to the areas that need them most. The array was calculated through layout optimization. Compared with a conventional microchannel approach, heat-transfer effectiveness increased by four to seven times.
We also optimized the full thermal path: chip hotspot → thermal interface → heat spreader → exchange surface → coolant. The base uses a diamond–silicon-carbide composite heat spreader. Diamond, with thermal conductivity of 2,200 W/(m·K), spreads heat laterally, while the SiC matrix provides mechanical strength and manufacturability. Base roughness is controlled below Ra 0.4 μm because contact resistance directly affects final junction temperature.
