Mains-connected computational resources have been powered primarily by 12-V distribution feeds for decades. The 12-V distribution potential has been a convenient choice because it lets designers route power efficiently throughout small-to-medium-sized systems and operate inexpensive, off-the-shelf fans and disk drives directly from the distribution voltage.

Advances in the semiconductor fabrication process have driven significant reductions in critical dimensions, and system functional densities have risen substantially over the last decade. There has been a corresponding increase in load-power density along with a need for greater operating efficiencies across a broad range of mains-powered applications.

It became apparent that racks in server farms were approaching the practical limit for 12-V power distribution, so the industry started to embrace 48-V power buses as an economical and energy-efficient alternative. Although these large applications were perhaps the first to switch to 48-V distribution, they are by no means the only area that can benefit from the higher distribution potential.

Perhaps the most visible place 48-V distribution provides benefits is in long runs for heavy loads. Moving from 12 to 48-V runs reduces bus losses to one-sixteenth that of 12-V systems using the same sized bus bar. That’s a 94% reduction in power dissipated by the physical distribution network. And improvement of system efficiency reduces the load on the cooling subsystem.

With such leverage on bus losses, designers can often opt to accept somewhat more loss in exchange for reducing the cross-section of the distribution bus. For example, if reducing distribution losses by only 87% is acceptable, you can halve the size of your copper distribution feeds. Eventually, the law of diminishing returns limits these gains. Dissipation in distribution feeds results from I²R losses. A factor increase in voltage roughly reflects as the reciprocal in current, so the square-law performance curve gives a lot of leverage and flexibility.

For large server-farm operators, switching to 48-V in-rack distribution brings benefits beyond reducing distribution losses. It has facilitated switching to 400-V dc plant power, which brings its own advantages compared to three-phase ac mains. For many head-end power-management blocks, conversion to 48 V is more efficient than to 12 V. Additionally, innovations in low-voltage power management have made possible direct 48-V-to-POL blocks that also exhibit high efficiency and impressive power densities.

You don’t have to be building 10-kW server racks to take advantage of 48-V power distribution, but the technology isn’t for flea-power applications either. Complicating the perspective somewhat, the power point at which 48-V distribution does make sense appears to be falling for a few reasons:

• Simple economies of scale are making the power-conversion blocks more affordable as adoption into new applications ramps manufacturing run rates.
• New, more cost-sensitive applications are either adopting or are poised to adopt the technology which has prompted component manufacturers to squeeze cost out of the building blocks to capture the growing and elastic market.
• The development of 48-V-to-POL reference designs reduces design risk and provides good performance data for comparisons with other approaches, particularly with regard to power density and total power-train cost.

If 48 V is a good potential for in-rack or in-box power distribution, why not go still higher to, say, 96 V? Surely, the benefits keep accruing. That’s true, if the only criterion that mattered were distribution losses. But other considerations suggest that 48-V isn’t just a good idea but rather, for many applications, it may well be close to an optimal choice of distribution voltage in mains-connected systems.

Power-management component manufacturers have produced efficient, dense, and economical blocks for both the head end — with either ac mains or HVDC sources — and for load-side conversion, either to a common board-level supply-rail potential or directly to the POL. Energy efficiency and power density would likely fall off with another doubling of distribution voltage.

Another, perhaps more important consideration that might limit moving to still-higher distribution potentials is that a 48-V nominal rail, with a +20% over-voltage protection limit still falls within the compliance requirement of 60 V max for SELV (safety extra-low voltage) circuits. Staying within SELV limits simplifies mechanical designs and permits deployment of low-cost low-voltage connectors. It also allows manufacturers and service providers to employ readily available assembly workers and service technicians without the additional cost and training time required for high-voltage qualification.