GaN Talk Blog

Michael A. de Rooij and Alejandro .P. Pozo

This article orginally appeared on EEPower.

Artificial intelligence is now in vogue - and its influence will only continue to grow. Today, the real challenge lies in delivering the energy demands at optimal efficiency levels. NVIDIA recently released a white paper¹ showing that an 800‑volt DC architecture combined with energy storage solutions represents an alternative approach to meeting the needs of modern and future AI server infrastructures. These systems will require power delivery in megawatts, not just kilowatts. The enabling technology is GaN. In collaboration with NVIDIA, EPC has developed a cost-effective, low-profile 800 VDC‑to‑12.5 VDC converter² capable of delivering up to 6 kW. The design leverages EPC’s latest 150 V and 40 V GaN device^3,4, fitting within a footprint of 5,000 mm² and a total thickness of only 8 mm. The design exemplifies how GaN FETs can achieve a combination of high efficiency and high power-density while keeping costs low, aiming to achieve 97% efficiency at full load.

There is increasing demand for extracting more power from standard 48 V bus converters for server applications' ever-increasing power requirements. GaN power devices allow a designer to achieve a converter's goals: high efficiency, small size & high current handling, ease, and high reliability. To whet your appetite, EPC designed a 1.2 kW resonant converter demo board (EPC9174) in an 1/8^th brick form factor that achieves an impressive 97.3% peak efficiency.