计算机科学
工程类
电气工程
工作(物理)
汽车工程
过程(计算)
功率(物理)
生产(经济)
自动化
管道(软件)
作者
Pranav Raj Prakash,Qiang Li
出处
期刊:Conference proceedings
[Institute of Electrical and Electronics Engineers]
日期:2026-03-22
卷期号:: 2790-2797
被引量:3
标识
DOI:10.1109/apec51134.2026.11517167
摘要
The rapid progress in artificial intelligence (AI) and machine learning (ML) has driven the development of next-generation, high-performance graphical processing units (GPUs), with multiple devices integrated on a single compute tray to minimize latency. This trend significantly raises the power demand of individual server racks and, consequently, entire datacenters. The conventional 48 V bus architecture is no longer sufficient, as it requires larger copper bus bars and occupies excessive space within the rack for power processing. To overcome these limitations, high-voltage DC (HVDC) bus architectures operating at ±400 V and 800 V have been proposed, thereby reducing copper volume and improving rack space utilization. Realizing these architectures requires compact, highly efficient 800 V/50 V unregulated DC–DC converters mounted on compute trays. This paper presents the design and implementation of a 6 kW GaN-based LLC converter designed for this application. The prototype achieves a peak efficiency of 98.75% and delivers a power density of 2070 W/in3, demonstrating its suitability for next-generation datacenter power delivery.
科研通智能强力驱动
Strongly Powered by AbleSci AI