Infineon looks to 2000A for AI power module
Infineon Technologies has developed dual-phase power modules as a key step to improving the efficiency of power supplies for AI in data centres.
The TDM2254xD series dual-phase AI power modules developed by Infineon use OptiMOS MOSFETs use novel packaging and proprietary magnetic structure for higher electrical and thermal performance.
The design of the module allows for efficient heat transfer from the power stage on to the heat sink through novel inductor design that is optimized to transfer current and heat. The modules support peak currents of up to 160A in 10x9x8 mm and 10x9x5 mm for 160 A peak, with the small footprint enabling vertical power delivery, reducing power delivery losses and increasing power density. This gives an efficiency two percentage points higher at full load to 89% and operating temperature 5°C cooler at full load.
- AI drives up data centre power requirements
- Blueprint for optimizing data centres for AI power
- Power Integrations teams for AI power supply design
Artificial Intelligence is currently driving an exponential increase in global data generation, and consequently increasing the energy demands of the chips supporting this data growth.
GPU-based AI servers require three times more energy than traditional servers, and data centres already consume more than 2 percent of the global energy supply, so it is essential to find innovative power solutions and architecture designs that further drive decarbonization.
Dual-phase power modules are critical to meet the power demands of high-power GPU systems by providing high power density and best in class efficiency without compromising signal integrity. By locating modules close to the processor and reducing power delivery losses, achieving >2000 A becomes feasible says the company.
Improving power efficiency at the core yields significant energy savings at scale. Since typical deployments consist of ~100,000 processors, saving 130 W/processor translates into megawatts for a datacentre.
Improving power efficiency at the core of a GPU yields significant energy savings at scale. This translates into megawatts saved for data centres computing generative AI and in turn leads to reduced CO 2 emissions and millions of dollars in operating cost savings over the system’s lifetime.
The TDM2254xD dual-phase power modules combine with Infineon’s XDP power controller to enable efficient voltage regulation for high-performance computing platforms with superior electrical, thermal and mechanical operation.
The 2MHz modules have 70 A per stage thermally managed with 120 A per phase peak, 4.25 V to 16 V input and 0.225 V to 3 V output. Current sensing and reporting is key for data centre AI, and the module reports data between at 8 uA/A to 8 mV/°C.
Infineon introduced the TDM2254xD series at the Applied Power Electronics Conference (APEC) in the US today.
“This unique Product-to-System solution combined with our cutting-edge manufacturing lets Infineon deliver solutions with differentiated performance and quality at scale, thereby significantly reducing total cost of ownership for our customers,” said Athar Zaidi, Senior Vice President, Power & Sensor Systems at Infineon Technologies. “We are excited to bring this solution to market; it will accelerate computing performance and will further drive our mission of digitalization and decarbonization.”
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
