Liquid Cooling And GaN: A Winning Combination

Data centers are facing an unprecedented transformation due to the surge in generative AI and other emerging technologies. A single ChatGPT session consumes 50 to 100 times more energy than a comparable Google search, escalating data center rack power requirements towards 200 kW or more, presenting serious challenges for operators.

Cooling, in fact, takes up about 40% of the power requirements that traditional air-cooling methods struggle to handle, prompting a shift toward liquid cooling. Even a partial (75%) transition to liquid cooling can reduce facility power consumption by up to 27%. This also reduces carbon emission by the same amount. Ironically enough, liquid cooling actually ends up using less water than air-cooling.

While liquid cooling represents a promising step in addressing the growing heat management and efficiency concerns, it solves only part of the problem.

Rack density and the power supply challenge

Efficient power delivery for CPUs and GPUs depends on the effectiveness of AC-DC power supplies. Silicon-based AC-DC converters are prone to higher switching losses, heat generation, and limited power density. These inefficiencies result in significant energy losses within data centers, driving up costs and emissions. Furthermore, with great power demands come great power density requirements. Existing converters, due to their limited power density, end up taking valuable real estate.

GaN power semiconductors offer smaller, faster, and more efficient solutions

Infineon’s CoolGaN solutions equipped with a commutation-robust fast-body diode with zero reverse recovery (Q_rr) offer fast, nearly lossless switching and the highest power density. This leads to smaller packages with the highest efficiency (over 99% efficiency for PFC topologies).

Fig. 1: Global data center energy saving potential with CoolGaN devices.

For every 10 racks in a datacenter, switching to GaN can increase savings by 3 million USD and decrease carbon emissions by 100 tons every year.

Liquid cooling favors GaN over SiC

Power designers are selecting GaN and SiC devices for their solutions. Interestingly, GaN solutions when paired with cold-plate liquid cooling result in lower operating temperatures for the server and power supply electronics.

Fig. 2: Increasing GaN performance with decreasing operating temperature.

GaN emerges as the clear frontrunner at these lower operating temperatures, outperforming SiC substantially in terms of losses connected to switching speeds and parasitic capacitance. With the improved power density offered by GaN, data centers can meet the higher power-per-rack requirements essential for AI workload demands.

Conclusion

Combining liquid cooling technologies and GaN semiconductors presents a massive opportunity for data centers to maximize efficiency, address rising power demands, and overcome the challenges posed by increasing heat. Industry leaders must take advantage of this moment to drive change, ensuring a future where data centers meet and exceed the expectations of a rapidly evolving digital landscape.