12kW AI power supply for 120kW datacentre racks
Navitas Semiconductor has developed a 12kW power supply unit (PSU) reference design that is designed for production for hyperscale AI datacentres.
The 12kW PSU combines gallium nitride (GaN) and silicon carbide (SiC) devices and complies with Open Rack v3 (ORv3) specifications and Open Compute Project (OCP) guidelines for 120kW racks.
The PSU is 790 x 73.5 x 40 mm and has an input voltage range of 180 – 305 VAC, outputting up to 50 VDC and delivering 12 kW when using input voltages above 207 VAC and 10 kW below this.
Nvidia consortium pushes 800V power distribution for AI datacentres
Navitas is also one of six semiconductor suppliers to Nvidia for its next generation of high power rack designs based on a 800V power distribution architecture. Another of these suppliers, Infineon, already has a 12kW reference design for AI datacentre power supplies using GaN and SiC devices.
The Gen-3 Fast trench-assisted planar SiC MOSFETs and high-power GaNSafe ICs are configured in 3-phase interleaved totem pole power factor correction (TP-PFC) and full bridge LLC topologies respectively along with the Navitas IntelliWeave digital control software.
IntelliWeave provides a hybrid control strategy of both Critical Conduction Mode (CrCM) and Continuous Conduction Mode (CCM), for light-load to full-load conditions, ensuring maximum efficiency while maintaining a simplistic design with low component count. This results in a 30% reduction in power losses compared to existing Continuous Conduction Mode (CCM) solutions.
The 3-phase interleaved full-bridge (FB) LLC topology integrates control, drive, sensing, and critical protection features with short-circuit protection with a 350ns maximum latency, 2kV ESD protection on all pins, elimination of negative gate drive and programmable slew rate control.
All these features are controlled with four pins, allowing the package to be treated like a discrete GaN FET, requiring no VCC pin. Suitable for applications from 1 kW to 22 kW, 650 V GaNSafe in TOLL and TOLT packages are available with a range of RDS(ON)typ. from 18 to 70 mΩ.
Additionally, it implements active current sharing and over-current, over-voltage, under-voltage, and over-temperature protections. It has an operating temperature range of -5 to 45oC, a hold-up time of ≥20 ms at 12 kW, and an inrush current of ≤3 times the steady-state current below 20 ms. Cooling is via the PSU’s internal fan.
“The continuation and leadership of Navitas’ AI power roadmap has seen a quadrupling in output power – from 2.7 to 12kW – in just over 24 months,” said Gene Sheridan, CEO and co-founder of Navitas.
“This increase in power delivery is vital for the world’s data centres to support the exponential power demanded by the latest GPU architectures. The ‘designed for production’ PSU enables our customers to quickly implement a highly efficient, simple, and cost-effective solution to address the power delivery challenges for AI and hyperscale data centers.”
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
