For the PDF version of this article, click here.
The distributed power architectures used in communications and computing typically rely on board-mounted power converters to generate the voltages required at the load. These on-board dc-dc converters take a 48-V supply that's bussed to each card in the system and step it down to create all of the necessary lower voltages. A common approach has long been to use a single-stage of dc-dc conversion on board with isolated dc-dc converters (“bricks”) generating all the voltages (see part a of the figure).
But with the number of supply voltages on the board rising, designers are now opting for a two-stage approach to dc-dc conversion. Using an isolated converter, the 48-V bus is converted into an intermediate voltage bus, typically at 12 V, 5 V or some value in between. That voltage, in turn, is stepped down by a series of nonisolated pointed-of-load buck converters, often referred to simply as POLs. Designated the intermediate bus architecture (IBA), this approach reduces power system cost by paying for isolation just once (see part b of the figure).
Although introducing an extra stage of dc-dc conversion can degrade efficiency, this can be mitigated by modifying the isolated converter in two ways. One is by loosening the converter's regulation requirements since the POLs are regulated. Another is by restricting the converter's input voltage range — something that's unacceptable in battery-backed telecom applications but appropriate in computing applications. Converters built with one or both of these modifications are labeled “bus converters” to distinguish them from the more-established bricks, although they often share the brick packaging formats.
The IBA represents an evolutionary change in distributed power design. The idea of mixing isolated and nonisolated converters on their boards is not new. Neither is the idea of two-stage on-board power conversion. Similarly, bus converters represent an easing of converter design specs, rather than a major redesign. Nevertheless, IBA still represents a modest repartitioning of the voltage conversion, isolation and regulation functions.
A more radical attempt to repartition these functions was introduced about a year ago when Vicor announced its Factorized Power Architecture (FPA). This architecture also relied on two-stage power conversion, but with the first stage providing just regulation at a relatively high voltage, such as 48 V, and a second stage offering voltage conversion and isolation at the point-of-load.
To implement this strategy, the company has developed a series of BGA-style modules that exploit a novel high-frequency circuit topology (sine amplitude converters). The main objective of this approach is to provide significantly higher current/power densities and efficiencies than are possible with IBA or other brick-based forms of DPA.
Naturally, the more aggressive repartitioning of power conversion functions by FPA forces designers to address more fundamental issues in power system design. For example, is the high-voltage bus fed to the board appropriate for distribution on the board? Also, what degree of regulation is required at each load?
Meanwhile, much of the power supply and power component industry labors to address challenges associated with the IBA. Since 2002, when discussions about IBA began heating up, the number of POLs introduced to the marketplace has exploded. In part, this is because so many brick vendors now offer families of POLs in various SIP and surface-mount-packages. But it's also the case that these POL product families must encompass many models to handle the various input and output voltage combinations at the various current levels. The lack of standardization in mechanical and electrical specs among the various offerings has been a source of confusion.
Several efforts are underway to simplify POL product selection. For instance, there are multivendor collaborations in the form of second-source and open standards agreements. Then, too, programmable POLs are emerging. These devices, which may be built with either digital PWMs or microcontroller-based configurability, are helping to make a single POL design more flexible in terms of its input and output voltage options.
At the same time, the programmability of these devices enables them to perform a variety of tasks, such as supply sequencing, monitoring, margining and adjustable fault protection. Whether achievable or not, the ultimate goal seems to be a “one-POL-fits-all” nonisolated converter.
In terms of POL development, the module makers are competing more or less head-to-head with power semiconductor vendors who offer dc-dc converters with various levels of integration, achieved monolithically in silicon or through multichip packaging.
At the same time, the semiconductor makers are working hard to tempt power supply vendors and OEMs alike with simple chipset solutions for isolated dc-dc converters. (The chip makers walk a tricky line in developing products that both cater to and compete with the power supply makers.) This embedded approach to implementing the isolated converter promises more of what IBA offers — cost reduction.
As the chip makers continue to develop isolated converter designs, they have opportunities to integrate functions beyond that of the standard brick. One is hot-swap control, a feature that may be appearing soon in bricks. Other opportunities lie in the integration of the board-level power management functions within the isolated converter — something brick vendors have talked about.
As these and other potential advances unfold, there may be further attempts to “divide and conquer” the various power-conversion functions on the board. These changes may force us to redraw the popular distributed power architectures once more.
Robert White, “Emerging On-Board Power Architectures,” Artesyn Technologies, APEC 2003 proceedings, pp.799-804.
For more on FPA, see Vicor's “Power Industry Articles” Web page at www.vicor.com/library/power_resources/industry_articles/.
The list of companies that have developed bus converters or POLs includes Artesyn Technologies, Astec Power, Bel Fuse, C&D Technologies, Celestica, Cherokee, Datel, Ericsson, Power One, SynQor, Texas Instruments and Tyco Electronics.
For more information on this article, CIRCLE 346 on Reader Service Card