One of the greatest challenges and opportunities facing experts in power management is the delivery of an increasingly specialised blend of current required to power high performance processing inside notebooks, desktops, and servers. Over the past decade, current levels have skyrocketed while Voltage input levels have plunged as we try to squeeze more work from platforms approaching their thermal breakpoint. In order to extend advances in computing performance, a systems approach to power management is required that boosts power efficiency and reduces power losses.
TRADITIONAL DESIGN APPROACH A typical power architecture applied to high-end computing is illustrated in Figure 1. It's a 2kW converter found within server or communications infrastructure for central office switchers, base stations or long-haul network switching stations.
There are four stages of power conversion. The AC input (stage 1) continues through a power factor correction and bulk DC-DC stage (stage 2) that reduces 380 Volts down to 48 Volts typically. In stage 3, the current passes through isolated DC-DC converters, often called bricks, half bricks, or fractional bricks. These 'bricks' live in many different places in the system, sometimes on the circuit board, sometimes in the rack. They convert the power again, usually to 3.3 or 12Volts, as in this example. At stage 4, the current is passed along to the many point-of-load converters which live right on the PC board, typically near the load in a variety of Voltages.
At each of these stages described, cost is added while performance is lost. At the power factor correction and bulk DC/DC conversion stage, often a critical conduction mode PFC circuit is used followed by a full bridge circuit, possibly zero Voltage switching which employs a 48Volt output rectification circuit using 200 to 300Volt fast recovery output diodes. Given current technology's performance in this configuration, typical full load efficiencies are in the range of 90%, the result of the 97% loss in the power factor boost converter and the 93% loss in the bridge plus output rectification circuit.
At the next stage, where isolated DC-DC converters (or bricks) are employed, our example requires at least ten bricks. Using a 12Volt rail, bricks with a 3.3Volt output are used often depending on system architecture. Alternatively, some designs convert all the way down to 3.3Volts and distribute higher currents at 12Volts, to lower distribution losses. Either way, full load efficiencies are in the range of 93% and the cost of this system is typically in the range of 20 to 50%/Watt. These bricks are highly complex and expensive systems with very well regulated outputs.
At the point of load stage, existing point of load converters usually are in a buck topology. Oftentimes, multiphase buck converters are selected to deal with exponentially increasing high currents at lower and lower Voltages. These multiphase converters add significant cost to the system and struggle to maintain 80% efficiency as the Voltage drops and the current increases. For our typical 2kW example, the total average full load efficiency is 82% across all points of load.
Taken in aggregate, this traditional system described yields an unimpressive 68% efficiency, requiring about 3kW in, to yield 2kW out, and carries a $1,000 price tag. By applying a systems perspective, much can be done at each stage to improve overall efficiency and cost.
IMPROVED SYSTEMS APPROACH First, let's look at the point of load converters. Traditional Voltage regulation module designs (eg VRM 9.1) can deliver up to 100A of power with 7–82% efficiency. These multiphase architectures are often limited to four phases and are quickly approaching the end of their usefulness as 100A becomes the minimum level for newer generations (VR 10.1 and 10.2).New power management architectures, like International Rectifier's XPhase, allow the number of phases to be scaled upward from one to X number of phases as design requirements expand. When combined with heat-dissipating MOSFET packages technologies that allow for both topside and bottom-side cooling, a system efficiency of 88% can be achieved for the example 2kW system.
Still greater cost and efficiency gains can be realised through major architectural changes in the next stage, the isolated DC-DC converters. With a new approach to the DC bus transformer architecture, the design can be simplified dramatically, boosting efficiency while stepping down the PFC-rectified line from 380 Volts, directly down to the 8–12 Volt rail. This completely eliminates the need for the secondary 48-Volt conversion and its associated distribution losses. This new architecture in IR's DCbus family is an open-loop unregulated converter, but the regulation of the 380Volts out of the PFC and the high efficiency of the follow-on buck converters, make this a very attractive and workable option. In the 2kW system, four parallel, 500Watt bulk DC-DC converters would be used. When combined with highly efficiency MOSFETs in synchronous rectification typical full load efficiencies would be 93%.
In the PFC section of Figure 1, high performance power factor correction circuits are accomplished only through very complex and critical conduction mode circuits using about 52 components. IR's new proprietary single-cycle control cuts component count by 75% while delivering a power factor of .999 and a total harmonic distortion of 2.5%.
The new approach, illustrated in Figure 2, yields a more efficient approach to DC-DC power conversion, at half the cost and half the power losses. Systems approaches like these will quell the flames now threatening next-generation processors, extending our digital roadmap.
