The surge in demand for artificial intelligence (AI) and high-performance computing (HPC) is driving unprecedented advancements in power supply design for data centers. Navitas Semiconductor has unveiled its groundbreaking 4.5kW AI data center power supply reference design, which sets new industry standards for power density and efficiency, addressing the towering needs of next-generation AI GPUs. As AI processors, like NVIDIA’s Blackwell series, evolve to require more significant power, the need for innovative power solutions becomes crucial. Navitas Semiconductor meets this challenge head-on with disruptive GaN and SiC technology.
This new 4.5kW power supply design from Navitas is noted for achieving an extraordinary power density of 137W/in³ and over 97% efficiency. These milestones have been attained by leveraging cutting-edge Gallium Nitride (GaN) and Silicon Carbide (SiC) semiconductor technologies. These advanced technologies serve as the backbone of the design, facilitating high efficiency and compact form factors that are vital in modern AI data centers. The utilization of GaN and SiC components is critical in enabling efficiency levels that far exceed conventional silicon-based approaches, which is a significant achievement given the steep energy demands of AI workloads.
Leading the Charge with 4.5kW Reference Design
Navitas Semiconductor’s latest 4.5kW power supply reference design has achieved an impressive power density of 137W/in³ and over 97% efficiency. These milestones are made possible by leveraging cutting-edge Gallium Nitride (GaN) and Silicon Carbide (SiC) semiconductor technologies. These technologies are at the core of the design, enabling high efficiency and compact form factors critical for modern AI data centers.
The design prominently features SiC components in the interleaved continuous conduction mode (CCM) totem-pole power factor correction (PFC) stage and GaN technology in the full-bridge LLC topology. The seamless integration of these advanced semiconductors ensures the power supply not only meets but surpasses current industry efficiency standards, including compliance with ‘Titanium Plus’ benchmarks.
Navitas’s high power density achievements are not just a matter of technological prowess but also an industry necessity. Data centers are looking to maximize their space utilization while handling higher power loads than ever before. This calls for power solutions that can offer high performance within smaller footprints. Hence, the 4.5kW design by Navitas is revolutionary in both its power efficiency and spatial economy. This development translates directly into enhanced operational capabilities for data center operators, allowing for more efficient configuration and deployment of high-power-demand AI servers.
Addressing Rising Power Demands
The escalating power demands driven by next-generation AI GPUs such as NVIDIA’s Blackwell B100 and B200, which require over 1kW each, have compelled a significant increase in power-per-rack specifications in data centers. Previously, data centers operated within a 30-40kW range per rack, but contemporary needs are pushing this figure to as high as 100kW.
Navitas Semiconductor’s power supply solutions are at the forefront of addressing these demands. By achieving higher power densities, Navitas not only allows for more efficient use of space but also contributes to the sustainable operational models increasingly sought after in data center environments. Enhanced efficiency translates directly into reduced CO2 emissions, aligning with global sustainability goals.
Sustainability and environmental impact are becoming crucial considerations in the data center industry. Efficient power supplies such as Navitas’s 4.5kW design are pivotal in reducing the carbon footprint of data centers. Beyond reduced space requirements, the improved efficiency of power conversion minimizes energy losses, resulting in lower operating costs and fewer emissions. This focus on sustainability does not come at the expense of performance, ensuring that data centers remain capable of supporting the intensive computational tasks demanded by AI and HPC applications.
Innovations in Semiconductor Technology
Central to Navitas’s breakthrough 4.5kW design are key innovations in semiconductor technology. The use of 650V Gen-3 ‘Fast’ (G3F) SiC MOSFETs that feature trench-assisted planar technology is a game-changer. This advancement ensures exceptional performance even under varying temperature conditions, significantly enhancing system reliability and efficiency.
Additionally, 650V GaNSafe power ICs in the LLC stage stand out due to their integrated design, which combines power, protection, control, and drive capabilities within a robust TOLL power package. These ICs exhibit low switching losses and can handle transient voltages up to 800V. High-speed benefits include low gate charge (Qg), minimal output capacitance (COSS), and the absence of reverse-recovery loss (Qrr), collectively elevating the system’s overall performance.
These semiconductor innovations are not just theoretical advancements; they are practical solutions that address the real-world challenges faced by modern data centers. High-speed switching is particularly crucial as it allows the reduction of passive power supply component sizes, such as transformers and capacitors. This not only decreases the overall size and weight of the power system but also reduces costs and improves reliability. The ability of these semiconductors to withstand high temperatures and voltages ensures robust performance in demanding environments, making Navitas’s design a versatile solution for varied data center applications.
Sustainability and Efficiency in Design
Navitas’s commitment to sustainability is evident in the high power densities achieved, which contribute to dematerialization—reducing the amount of physical material needed for components. High-speed switching reduces the size, weight, and cost of essential power supply elements such as transformers, capacitors, and EMI filters.
Optimizing the design’s power efficiency also translates into significant sustainability benefits. Increased system efficiency reduces overall energy consumption and thereby decreases CO2 emissions, marking a substantial ecological footprint improvement while simultaneously catering to the pressing energy requirements of the data center industry.
The 4.5kW design’s attention to sustainability does not merely add environmental credentials but also addresses practical operational concerns. For instance, the reduction in the size and weight of power supply components directly impacts the cooling requirements of data centers. Smaller and more efficient components generate less heat, thus reducing the burden on cooling systems, which in turn saves energy and lowers operational costs. This holistic approach to design underscores Navitas’s commitment to not just meeting but exceeding the industry’s expectations in both performance and sustainability metrics.
Market Reception and Future Outlook
The surge in demand for artificial intelligence (AI) and high-performance computing (HPC) is fueling significant advancements in power supply design for data centers. Navitas Semiconductor has introduced a revolutionary 4.5kW AI data center power supply reference design, setting new benchmarks for power density and efficiency to meet the towering demands of next-generation AI GPUs. As AI processors, such as NVIDIA’s Blackwell series, evolve and require more substantial power, the need for innovative power solutions becomes increasingly crucial. Navitas Semiconductor rises to this challenge with groundbreaking GaN and SiC technology.
This new 4.5kW power supply design from Navitas is celebrated for its remarkable power density of 137W/in³ and its exceptional efficiency of over 97%. These achievements are made possible through advanced Gallium Nitride (GaN) and Silicon Carbide (SiC) semiconductor technologies. These advanced components are crucial, enabling high efficiency and compact design, essential for today’s AI data centers. The use of GaN and SiC components significantly surpasses the efficiency levels of conventional silicon-based methods, an impressive feat given the steep energy demands of AI workloads.