World’s First AI Power Supply With GaN And SiC

The world’s first 8.5kW power supply for AI data centres offers 98% efficiency and new technology. Can it address the industry’s growing power demands? Find out!

Navitas Semiconductor has introduced the world’s first 8.5 kW power supply unit (PSU) that uses GaN and SiC technologies to achieve 98% efficiency. Designed for AI and hyperscale data centres, this PSU offers fast switching and robustness, making it suitable for various applications, including EV charging.

The 54V output PSU meets Open Compute Project (OCP) and Open Rack v3 (ORv3) standards. It uses GaNSafe and Gen-3 Fast SiC MOSFETs in a 3-phase setup to ensure high efficiency and performance with fewer components. The 3-phase design, compared to the 2-phase design, reduces ripple current and EMI compared to other PSUs. The PSU also reduces the number of GaN and SiC devices by 25% compared to competitors, lowering overall costs.

– Advertisement –

The PSU operates with an input voltage range of 180 to 264 Vac, provides a 12V standby output, and has an operating temperature range from -5°C to 45°C. Its hold-up time is 10 ms at 8.5 kW, which can be extended to 20 ms with an extender. These features make the PSU suitable for demanding applications in environments that require both high power and reliability.

Powered by GaNSafe, the PSU offers advanced protection features, including short-circuit protection with a maximum latency of 350ns and 2kV ESD protection on all pins. It also eliminates negative gate drive and includes programmable slew rate control, all controlled via four pins. GaNSafe is suitable for applications ranging from 1 kW to 22 kW and is available in 650V TOLL and TOLT packages.

“This complete wide bandgap solution of GaN and SiC enables the continuation of Navitas’ AI power roadmap, which enables this 8.5kW and plans to drive to 12kW & higher in the near term”, said Gene Sheridan, CEO and co-founder of Navitas. “As many as 95% of the world’s data centres cannot support the power demands of servers running NVIDIA’s latest Blackwell GPUs, highlighting a readiness gap in the ecosystem. This PSU design addresses these challenges for AI and hyperscale data centres.”

For more information, click here.