New power-management solutions deliver scalable AI infrastructures

Texas Instruments has debuted new design resources and power-management chips to assist companies in meeting growing AI computing demands and scaling power-management architectures from 12V to 48V to 800VDC.

The company is collaborating with NVIDIA to develop power-management devices to support 800VDC power architecture, as IT rack power is expected to eclipse 1MW in the next two to three years. The company's white paper, 'Power delivery trade-offs when preparing for the next wave of AI computing growth', reexamines the power delivery architecture within the IT rack, and addresses design challenges and opportunities for high efficiency and high power-density energy conversion at a system level.

The company's reference design for a 30kW AI server power-supply unit supports stringent AI workloads. Its dual-stage power-supply reference design features a three-phase, three-level flying capacitor power factor correction converter paired with dual delta-delta three-phase inductor-inductor-capacitor converters. The power supply is configurable as a single 800V output or separate output supplies.

A dual-phase smart power stage has the highest peak power density power stage on the market. The CSD965203B offers 100A of peak current per phase and combines two power phases in a single 5mm x 5mm quad flat no-lead package. The device enables designers to increase phase count and power delivery within a compact printed circuit board area, thereby improving efficiency and performance.

The CSDM65295 module delivers up to 180A of peak output current in a compact 9mm x 10mm x 5mm package, helping engineers increase data centre power density without compromising thermal management. The module integrates two power stages and two inductors with trans-inductor voltage regulation (TLVR) options, while maintaining high efficiency and reliable operation.

The company's GaN intermediate bus converter is capable of delivering up to 1.6kW of output power in a quarter-brick (58.4mm x 36.8mm) form factor. The LMM104RM0 converter module offers over 97.5% input-to-output power-conversion efficiency and high light-load efficiency to allow active current sharing between multiple modules.

AI data centres require architectures designed with multiple foundational semiconductors for efficient power management, sensing and data conversion. With new design resources and a broad power-management portfolio, TI is working alongside data centre designers to implement a comprehensive approach that drives efficient, safe power management – from power generation at the grid to the fundamental logic gates of graphics processing units.

"With the growth of AI, data centres are evolving from simple server rooms to highly sophisticated power infrastructure hubs," said Chris Suchoski, sector general manager, Data Centres at TI. "Scalable power infrastructure and increased power efficiency are essential to meet these demands and drive future innovation. With devices from TI, designers can build innovative, next-generation solutions that are enabling the transition to 800VDC."