Navitas Semiconductor has released an 8.5 kW power supply unit (PSU) that utilizes GaN and SiC technologies and has achieved 98% efficiency for AI and hyperscale data centers.
The AI-optimized 54V output PSU meets Open Compute Project (OCP) and Open Rack v3 (ORv3) specifications. It incorporates GaNSafe and Gen-3 Fast SiC MOSFETs in 3-phase interleaved PFC and LLC topologies, delivering high efficiency and performance with reduced component count. The 3-phase topology for both PFC and LLC results in low ripple current and EMI. The design uses 25% fewer GaN and SiC devices compared to alternative systems, reducing costs. The PSU operates with input voltage from 180 to 264 Vac, provides 12V standby output, and functions between -5°C to 45°C. At 8.5 kW, it maintains a 10 ms hold-up time, extensible to 20 ms.
The 3-Phase LLC topology uses high-power GaNSafe, which is designed for high-power applications in AI data centers and industrial markets. The 4th generation technology integrates control, drive, sensing, and protection features. GaNSafe includes short-circuit protection (350ns max latency), 2kV ESD protection on all pins, eliminates negative gate drive, and offers programmable slew rate control. These features operate with 4-pins, allowing package use similar to discrete GaN FET without requiring a VCC pin. The 650V GaNSafe in TOLL and TOLT packages suits applications from 1 kW to 22 kW, with RDS(ON)MAX ranging from 25 to 98 mΩ.
The 3-Phase interleaved CCM TP-PFC uses Gen-3 Fast SiC MOSFETs with “trench-assisted planar” technology developed through ongoing SiC innovation. This technology delivers consistent performance across temperatures, supporting EV charging and AI data center applications.
The PSU addresses power infrastructure challenges data centers face when implementing advanced AI GPU systems. It will be exhibited at Electronica 2024 (Hall C 3, booth 129, November 12th–15th).
The PSU will be displayed for the first time at Electronica 2024 (Hall C 3, booth 129, November 12th– 15th).
Leave a Reply
You must be logged in to post a comment.