3D printing cold plates for AI datacentre cooling
Fabric8Labs has developed a cold plate for AI accelerators that maps the design to the areas of the chip that produces the most heat.
The company uses its Electrochemical Additive Manufacturing (ECAM) 3D metal printing to produce fin structures that direct coolant flow to thermal hot spots.
ECAM is a room temperature metal printing process that does not need post-processing, as many other metal 3D printing techniques require. As a result, stress-free copper structures with a resolution of 33 microns can be printed directly onto temperature-sensitive substrates such as PCBs, silicon, or existing metal components.
This enables customized cooling with parametrically optimized fins, delivering optimized thermal-hydraulic performance, uniform die temperature distribution, and thermal performance enhancements 48% better than traditional microchannel technologies.
- Pegatron certifies two phase direct-to-chip cooling
- Nvidia opens up its AI factory stack with Fujitsu, Qualcomm
The cooling plate is being used by server maker Wiwynn for a 3.5kW cooling plate, as well as an optimized double-sided cold plate for future high-power chips.
Wiwynn has worked with Wistron to be among the first to offer a liquid-cooled Nvidia GB300 NV72 rack system, working with nVent on the in-rack coolant distribution unit (CDU). It has also developed a server based on the latest AMD Instinct MI350 series GPU that boosts AI inference by 35x over the previous Instinct MI300X and also requires higher levels of cooling.
“Wiwynn is a leader in data centre IT infrastructure, and we are thrilled to support their mission with advanced cold plates enabled by ECAM,” said Ian Winfield, VP of Product & Applications at Fabric8Labs, which is backed by imec.xpand, TDK Ventures, Intel Capital and Schnieder Electric. “ECAM enables the cooling solution to be designed specifically for the underlying SoC power map, optimizing performance, efficiency, and device reliability.”
The technology is also being used by AEWIN for 3D micro-mesh boiler plates for networking equipment in AI datacentres
The micro-mesh boosts the thermal performance by 1.3 °C/100W compared to alternatives. The ECAM 3D micro-mesh designs increase surface area by more than 900%, with high performance structures that serve as a capillary network to continuously refresh coolant at the boiling interface, drastically improving heat dissipation.
“Our collaboration with AEWIN represents a significant step forward toward the future of thermal management. We are thrilled to support AEWIN by enabling them to achieve their sustainability targets and meet the growing power demands of advanced AI accelerators,” said Winfield.
“The exponential growth of data and Edge AI complexity requires the most advanced on-premises computing. Through our advanced system-level design, we are able to leverage Fabric8Labs’ ECAM technology to optimize solutions for high efficiency, power usage effectiveness, and reduced total cost of ownership,” said Dr. Liu, Director of the Advanced Technical Development Division at AEWIN Technologies.
www.fabric8labs.com; www.wiwynn.com; www.aewin.com
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
