The Changing Landscape of Data Centers

In recent years, data centers have faced increasing pressure to improve efficiency and sustainability while dealing with growing computational demands. As artificial intelligence (AI) and machine learning applications proliferate, traditional power architectures struggle to keep up. Enter the revolution spearheaded by ROHM and NVIDIA: the introduction of an 800V High-Voltage Direct Current (HVDC) architecture that promises to transform the way data centers are designed and operated.

ROHM's Contributions to the New Era

As a leader in power semiconductor technology, ROHM plays a crucial role in this transformation. Their vast portfolio includes silicon and wide-bandgap technologies, notably silicon carbide (SiC) and gallium nitride (GaN) devices. These technologies provide a strategic advantage for data center designers looking to balance cost, efficiency, and reliability.

One of the cornerstone components in this paradigm shift is the RY7P250BM, a 100V power MOSFET optimized for hot-swap circuits in 48V power systems. This advancement is vital for the new AI server environments that necessitate high availability and minimal downtime. Other hallmark features include its best-in-class Safe Operating Area (SOA) performance and ultra-low ON-resistance, contributing to reduced power loss and enhanced system reliability.

The Significance of 800V HVDC Architecture

Traditional data centers utilize a 54V rack power system that suffers from inefficiencies due to space limitations, conversion losses, and copper overload. The 800V HVDC architecture introduced by NVIDIA aims to tackle these issues head-on by converting 13.8kV AC from the grid directly into 800V DC — a monumental improvement over previous architectures.

ROHM's SiC MOSFETs are at the heart of this innovation, offering exceptional performance in high-voltage environments. Their reduced switching and conduction losses, along with superior thermal stability, make them ideal for the compact and high-density systems required in cutting-edge AI data centers.

The Role of GaN Technology

While SiC technology addresses high-voltage applications, ROHM’s GaN solutions under their EcoGaN brand step in to cover ranges from 100V to 650V. GaN technology is renowned for its superior breakdown field strength and fast switching capabilities. These characteristics make it particularly effective for smaller, more efficient power systems, further supporting the rigorous demands of modern AI applications.

Advanced SiC Modules for Centralized Power Systems

Beyond discrete devices, ROHM offers a plethora of high-power SiC modules, including the HSDIP20 series. These modules boast advanced thermal performance and scalability, optimized for 1200V applications in AC-DC and DC-DC converters. By facilitating efficient and scalable power conversion, these modules are pivotal to NVIDIA's vision of centralized power systems and can handle workloads in 800V busways and megawatt-scale rack configurations.

Conclusion: A Step Towards Sustainable AI

The collaboration between ROHM and NVIDIA marks a significant advancement in the quest for more efficient and sustainable data centers. By shifting to an 800V HVDC architecture, data centers can drastically reduce copper usage and minimize energy losses while simplifying power conversion processes. This technological leap not only prepares the infrastructure for the increasing demands of AI but also sets a precedent for future advancements in the field.

The journey towards sustainable and efficient AI-focused data centers is indeed a long one, but with these innovations, industry leaders like ROHM and NVIDIA underscore their commitment to paving the way for a more sustainable digital future.