-
With the increasing power requirements of Artificial Intelligence (AI) processors, server power supplies (PSUs) must deliver more and more power without exceeding the defined dimensions of the server racks. This is driven by a surge in energy demand of high-level GPUs, which could consume 2 kW and more per chip by the end of the decade.
These needs, as well as the emergence of increasingly demanding applications and the associated specific customer requirements have prompted Infineon Technologies AG to extend the development of SiC MOSFETs to voltages below 650 V. The company is now launching the new CoolSiC™ MOSFET 400 V family, which is based on the second generation (G2) CoolSiC technology introduced earlier this year.
The new MOSFET portfolio was specially developed for use in the AC/DC stage of AI servers, complementing Infineon’s recently announced PSU roadmap. The devices are also ideal for solar and energy storage systems (ESS), inverter motor control, industrial and auxiliary power supplies (SMPS) as well as solid-state circuit breakers for residential buildings.
“Infineon offers an extensive portfolio of high-performance MOSFETs and GaN transistors to meet the demanding design and space requirements of AI server power supplies”, said Richard Kuncic, Head of the Power Systems Business Line at Infineon. “We are committed to supporting our customers with advanced products such as the CoolSiC MOSFETs 400 V G2 to drive highest energy efficiency in advanced AI applications.”
The new family features ultra-low conduction and switching losses when compared to existing 650 V SiC and Si MOSFETs. Implemented in a multi-level PFC, the AC/DC stage of the AI Server PSU can attain a power density of more than 100 W/in³ and is proven to reach 99.5 percent efficiency.
This is an efficiency improvement of 0.3 percentage points over solutions using 650 V SiC MOSFETs. In addition, the system solution for AI Server PSUs is completed by implementing CoolGaN™ transistors in the DC/DC stage. With this combination of high-performance MOSFETs and transistors, the power supply can deliver more than 8 kW with an increase in power density by a factor of more than 3 compared to current solutions.
The new MOSFET portfolio comprises a total of 10 products: five R DS(on) classes from 11 to 45 mΩ in Kelvin-source TOLL and D²PAK-7 packages with .XT package interconnect technology. The drain-source breakdown voltage of 400 V at T vj = 25°C. makes them ideal for use in 2- and 3-level converters and for synchronous rectification.
The components offer high robustness under harsh switching conditions and are 100 percent avalanche tested. The highly robust CoolSiC technology in combination with the .XT interconnect technology enables the devices to cope with power peaks and transients caused by sudden changes in the power requirements of the AI processor. Both the connection technology and a low and positive R DS(on) temperature coefficient enable excellent performance under operating conditions with higher junction temperatures.
Original – Infineon Technologies
-
The influence of artificial intelligence (AI) is driving up the energy demand of data centers across the globe. This growing demand underscores the need for efficient and reliable energy supply for servers. Infineon Technologies AG opens a new chapter in the energy supply domain for AI systems and unveils a roadmap of energy efficient power supply units (PSU) specifically designed to address the current and future energy need of AI data centers.
By introducing unprecedented PSU performance classes, Infineon enables cloud data center and AI server operators to reduce their energy consumption for system cooling. The innovative PSUs reduce power consumption and CO 2 emissions, resulting in lower lifetime operating costs. The powerful PSUs are not only used in future data centers but can also replace existing power supply units in servers and increase efficiency.
In addition to the current PSUs with an output of 3 kW and 3.3 kW available today, the new 8 kW and 12 kW PSUs will contribute to further increasing energy efficiency in future AI data centers. With the 12 kW reference board, Infineon will offer the world’s first power supply unit that achieves this level of performance and supplies future data centers with power.
“At Infineon, we power AI. We are addressing a critical question of our era – how to efficiently meet the escalating energy demands of data centers,” says Adam White, Division President Power & Sensor Systems at Infineon. “It’s a development that was only possible by Infineon’s expertise in integrating the three semiconductor materials silicon (Si), silicon carbide (SiC), and gallium nitride (GaN) into a single module. Our PSU portfolio is therefore not only an example of Infineon’s innovative strength, which leads to first-class results in terms of performance, efficiency and reliability for data centers and the AI ecosystem. It also reinforces Infineon’s market leadership in power semiconductors.”
Infineon is responding to the requirements of data center operators for higher system efficiency and lower downtimes. The growth of server and data center applications has led to an increase in power requirements, necessitating the development of power supplies with higher power ratings from 800 W up to 5.5 kW and beyond. This increase is driven by the growing power requirements of Graphic Process Units (GPU) on which AI applications are computed.
High-level GPUs now require up to 1 kW per chip reaching 2 kW and beyond by the end of the decade. This will lead to higher overall energy demand for data centers. Depending on the scenario, data centers will account for up to seven percent of global electricity consumption by 2030; this is an order of magnitude comparable to India’s current electricity consumption.
Infineon’s new PSUs contribute to the efforts to limit the CO 2 footprint of AI data centers despite the rapidly growing energy requirements. This is made possible by a particularly high level of efficiency that minimizes power losses. Infineon’s new generation PSUs achieve an efficiency of 97.5 percent and meet the most stringent performance requirements. The new 8 kW PSU is capable of supporting AI racks with an output of up to 300 kW and more. Efficiency and power density is increased to 100 watts per in³ compared to 32 W/in³ in the available 3 kW PSU, providing further benefits for the system size and cost savings for operators.
From a technical perspective, this is made possible by the unique combination of the three semiconductor materials Si, SiC and GaN. These technologies contribute to the sustainability and reliability of AI server and data center systems. Innovative semiconductors based on wide-bandgap materials such as SiC and GaN are the key to a conscious and efficient use of energy to drive decarbonization.
The 8 kW Power Supply Unit will be available in Q1 2025. For more information about the PSU roadmap, please click here.
Infineon at the PCIM Europe 2024
PCIM Europe will take place in Nuremberg, Germany, from 11 to 13 June 2024. Infineon will present its products and solutions for decarbonization and digitalization in hall 7, booths #470 and #169. Company representatives will also be giving several presentations at the accompanying PCIM Conference and Forums, followed by discussions with the speakers. Information about Infineon’s PCIM 2024 show highlights is available at www.infineon.com/pcim.
Original – Infineon Technologies
-
Navitas Semiconductor announced their AI data center technology roadmap for up to 3x power increase to support similar exponential growth in AI power demands expected in just the next 12-18 months.
Traditional CPUs require typically only 300W and the data center ac/dc power supplies would typically power the equivalent of 10 of these or 3,000W (3kW). High-performance AI processors like NVIDIA’s ‘Grace Hopper’ H100 are already demanding 700W each today, with next-gen ‘Blackwell’ B100 & B200 chips anticipated to increase to 1,000W or more by next year.
To meet this exponential power increase, Navitas is developing server power platforms which rapidly increase from 3kW to up to 10kW. In August 2023, Navitas introduced a 3.2kW data center power platform utilizing latest GaN technology enabling over 100W/in3 and over 96.5% efficiency. Now, Navitas is releasing a 4.5kW platform enabled by a combination of GaN and SiC to push densities over 130W/in3 and efficiencies over 97%. These two platforms have already generated significant market interest with over 20 data center customer projects in development expected to drive millions in GaN or SiC revenues starting this year.
Today, Navitas also announces its plans to introduce an 8-10kW power platform by the end of 2024 to support 2025 AI power requirements. The platform will utilize newer GaN and SiC technologies and further advances in architecture to set all-new industry standards in power density, efficiency and time-to-market. Navitas is already engaged with major data-center customers, with full platform launch anticipated in Q4 ’24, completing this 3x increase in power demands in only 12-18 months.
Navitas’ unique data-center design center is creating these system designs to address the dramatic increases in AI data center power requirements, and assist customers to deploy platforms quickly and effectively to meet the accelerated time-to-market demands of rapid AI advances. System designs include complete design collateral with fully-tested hardware, schematics, bill-of-materials, layout, simulation and hardware test results to maximize first-time-right designs and fast revenue generation.
“The rapid development and deployment of artificial intelligence (AI) into global data centers has created a dramatic and unexpected power challenge for our entire industry,” noted Gene Sheridan, Navitas’ CEO and Co-Founder. “Our investment in leading-edge GaN and SiC technologies, combined with our unique data-center design center capabilities, have positioned us well. Our team has really stepped up to the challenge, with a 3x power increase in less than 18 months.”
Original – Navitas Semiconductor
