Power supplies just may be surrounded by more myths and misconceptions than any other segment of electronics. Within the past year, the very nature of power-system design has changed, and the old rules have pretty much flown out the window. Although some of the following statements may once have been true, most of them, if not out-and-out false, are now accurate only in specialized niches of the power-supply field.

• Power supplies are low-tech.
• Except in unusual cases, any EE—and most electronics technicians—can design a suitable power supply; no special knowledge or experience is required.
• The power-supply field is a technological backwater, which, compared with the rest of the electronics industry, changes at a snail's pace.
• If you integrate power-supply modules into higher level products—as opposed to buying the parts and integrating the power-supply functions—you are wasting a lot of money. Most likely, you are doing so because power-supply vendors' scare tactics have taken you in.
• IC vendors' reference designs for supplies that you can build into your product are much less expensive than the equivalent modules. Moreover, because the IC vendor does all of the hard work, the built-in supply does not increase your development costs compared with those you would incur if you used purchased modules.

The facts are different. Power-supply design is only rarely trivial; it is a highly specialized craft that combines electronic, electrical, thermal, mechanical, and manufacturing engineering. Experienced power-supply designers—the good ones, anyhow—understand as do few others, the extremely complex and subtle trade-offs that can profoundly affect a design's reliability and cost. Even if you plan to integrate an IC manufacturer's power-supply reference design into your product, you would be well-advised to assign the task to someone experienced in power-supply design. Similarly, power-distribution experience can be a big help to engineers who must apply purchased power modules.

That said, the job of designing power supplies into higher level products greatly differs today from what it was even two or three years ago. The whole landscape of power-supply design has changed and continues to change in ways that strongly benefit designers of higher level products. Power supplies are dramatically smaller, more efficient, and less expensive than were their counterparts of just a short time ago. The cost reductions relate more to advances in technology than to the disastrous business conditions that currently plague the industry.

IC manufacturers—not power-supply designers—drove one of the major areas of change onto the power-supply industry. ICs are now designed to make optimum use of semiconductor-manufacturing processes that can support only limited supply voltages. Gone are large pc boards that used just a few supply voltages—often, 5V and ±12V. Today, such boards often use six or more supply voltages—for example, 5, 3.3, 2.5, 1.8, 1.5, and 1.2V, and sometimes –5V. Moreover, the voltages are now so low and the currents so high that IR drops in the distribution networks (that is, in the power and ground planes) can often make or break a design. And the voltages are going lower: 1, 0.8, 0.6, and even 0.5V are in the offing—generally with no decrease in the power the ICs require. Constant power at lower V_CC translates into higher I_C and the need for much lower R_D (distribution resistance).

Also, static (that is, I·R) drops are not the only voltage drops whose importance the new lower voltage ICs has magnified. Minimizing the power dissipated in those ICs has become ever more important—and not just in battery-powered products. Low power consumption is a goal even in large line-powered equipment, because reducing power dissipation reduces cooling requirements and lowers devices' operating temperature, thereby increasing equipment reliability.

Lowering IC power dissipation often dictates the use of power-saving modes. When they enter such modes, ICs demand much less supply current; when the devices return to "normal" operation, however, power supplies must suddenly deliver high currents. These drastic current changes produce voltage transients that the supplies usually must correct. (The formula e=L·di/dt describes the effect. In the formula, e is instantaneous voltage, L is the inductance between the supply and the load, and di/dt is the rate of change in the current delivered to the load.) Achieving tight regulation at the point of load increasingly requires supplies to incorporate remote sensing, a feature that, unless designed with the utmost care, can easily degrade a supply's transient response, dynamic stability, or both.

Transient-regulation requirements now often necessitate much more closely controlled dynamic behavior than that of the supplies system designers considered more than adequate just a few years ago. A key to improved dynamic behavior is higher switching frequencies. Most modern POL (point-of-load) converters use MOSFET switches. Without faster high-current switches, the higher switching frequencies would unacceptably limit efficiency.

In this regard, newer ICs' requirements for lower supply voltages have proved to be a boon; in general, increasing a MOSFETs' switching speed reduces both the devices' breakdown voltage and their on-resistance. Lower voltage supplies can tolerate lower breakdown-voltage switches. For any rated output current, lower on-resistance improves a converter's efficiency—even in designs that don't take advantage of the FETs' ability to switch faster.

New meanings

The mandate for multitudes of minuscule supply voltages has given new meaning to time-honored terms such as DPA (distributed-power architecture), IBA (intermediate-bus architecture), IBV (intermediate-bus voltage), and POL (for point-of-load regulation as well as point-of-load converter). BC, IBC, RBC, and BCM stand, respectively, for bus converter, intermediate-bus converter, regulated bus converter, and bus-converter module. Systems that use multiple supply voltages have also raised the importance of such features as supply sequencing, which not too long ago system designers could often safely overlook. Texas Instruments' PTHxx module family exemplifies the advanced sequencing capabilities that supply manufacturers are now starting to offer (Figure 1).

DPAs have existed for decades in a variety of forms. For several years, however—until about a year ago—many people understood the term to mean a DPA that originated in the telecom industry and had spread to other applications. This DPA, which remains popular, became the de facto standard for powering systems consisting of multiple boards, each of which usually consumes no more than approximately 100W. In this architecture, each board receives a relatively high IBV, usually a poorly regulated nominal 48V (actually, 36 to 75V).

In telecom and in some high-availability-computing applications, the 48V IBV usually originates either in a line-operated supply separate from the boards or—in the event of an ac-line failure—in a bank of large lead-acid storage batteries whose charge is normally maintained by the line-operated supplies. The wide 36 to 75V input range easily accommodates batteries in any usable state of charge. Because the conversion from ac-line voltage to IBV doesn't occur on the pc boards, some people don't even consider this arrangement an IBA.

Usually, one or more isolated, closely regulated "sub-brick" dc/dc converters located on the board, near the edge that contains the backplane connector, develop the lower dc voltages that power the board's ICs. Even though, unlike many POLs, these modules incorporate transformer-based input-to-output isolation and they are joined to the loads by pc-board traces that are often longer than 1 ft, these converters technically qualify as POLs.

Within the past year, however, a different DPA has been growing in popularity. This new IBA substitutes nonisolated POLs for the isolated sub-brick devices and concentrates the isolation in the BC. The BC is an isolated sub-brick module whose output is normally unregulated but in some cases is loosely regulated, for example to compensate for input-voltage variations but not for output-voltage variations that result from the BC's nonzero output impedance. The most significant advantage of this IBA is its inherently low cost (see sidebar, "It's the money, stupid!"). The most significant disadvantage is reduced efficiency. When you convert the voltage twice—from the voltage supplied to the board to the IBV and from the IBV to V_CC—you generally lose a couple of percentage points in efficiency. If your load draws 100W, the IBA typically consumes 2W more than would a single-conversion architecture. For many system designers, however, lower cost trumps lower efficiency.

Even with the cost-effective planar magnetics that most isolated sub-brick dc/dc converters now use, the transformer is a significant cost item. Manufacturers of dc/dc converters are in almost complete agreement that power systems that use only one isolation transformer are less expensive than systems that use many. The key exception to this view is Vicor, which recently introduced what it calls FPA (Factorized Power Architecture). This architecture uses individually isolated POLs, which the company calls VTMs (voltage-transformation modules), and separate regulators called PRMs (preregulation modules) (Reference 1). Usually, there is one PRM for each VTM. Vicor won't reveal how it achieves its favorable costs, but the company's initial releases suggest that FPA pricing is often lower than that of systems in which one BC powers several POLs. Moreover, the packaging, called VIC (V·I Chip), delivers exceptionally high power density, and the fixed clock frequency of almost 4 MHz should provide extraordinarily rapid transient response.

Vicor's FPA has raised many questions but so, too, has the aforementioned single-point-of-isolation IBA (see sidebar "Do you need an intermediate-voltage bus?"). Because lower cost was the main motivation for creating this IBA, some suppliers have ruled out certain choices that otherwise might make sense. For example, the BC might provide line regulation but not load regulation. A BC with regulation only at its input would allow the familiar 36 to 75V telecom input-voltage range and thus would facilitate battery backup. Also, eliminating the BC's ability to compensate for load changes would save the cost of analog isolation in the feedback path, thus making the BC less expensive than a fully regulated module. Still, if the power-supply manufacturer convinces the system manufacturer that there is no problem with limiting the BC's input-voltage range to, at most, ±10% from nominal, the POLs can do all of the regulation. Besides costing less than a unit with input regulation, an unregulated BC increases efficiency and power density and improves reliability.

Another controversy relates to the IBV. Although system designers have made a good case for IBVs of less than 12V, several power-module manufacturers believe that, except perhaps in a few high-volume custom applications, 12V must triumph because of the large infrastructure associated with it.

Module manufacturers say that they are hard-pressed to find customers who design their pc boards with the POLs in immediate proximity to the ICs they power, despite good reasons—such as improved transient response and possible elimination of remote sensing—for so locating the POLs. You might think that, with long ground-return paths and V_CC values in the vicinity of 1V, mixing different ICs' ground currents would create data-integrity problems, which could motivate designers to provide separate IBV sources either for each POL or for small groups of POLs. Such separate sources would facilitate single-point grounding. So far, though, system designers apparently haven't recognized a need for such schemes. Should the need arise, however, the IBA offers the possibility of a moderate-cost remedy. Power-supply manufacturers can design a BC that produces two isolated 10A outputs in the same-size package as a BC that produces one 20A output. The second output would add minimal cost even if the BC included regulation to compensate for input-voltage variations.

The $64 million (minimum) question is whether IBAs really hold the potential to wipe out the power-module industry. Power-IC manufacturers, of which you will find a partial list in the sidebar "For more information," believe that many companies that buy modular dc/dc converters are wasting their money and that these companies could save large amounts by using the IC manufacturers' reference designs to build their own converters—especially POLs.

Module manufacturers now agree that this statement is sometimes true—but not as often as most designers who work outside power-supply design believe. A module company cites the following experience with a large system manufacturer, which finds designing power supplies for use in its systems important enough for it to maintain a power-supply-design department. When the annual quantities of a design exceed 1 million units, and the supplies' output is less than approximately 6A or 20W, the customer designs and builds its own supplies as part of larger pc boards. The module maker says that the customer builds these supplies for approximately one-third of the price the module maker would have to charge.

Even though it has a power-supply-design department, the system manufacturer must see large cost differentials before it can justify the expense of developing a supply in house. This company buys supplies for which its annual requirements exceed 100,000 pieces—even when these supplies' output does not exceed the 6A/20W limit. In short, this system manufacturer has found that economics favor its building supplies only when the output power is modest and the quantities are large. If this company is representative of system manufacturers in general, the power-module business may never return to its former size, but it will remain sizable nevertheless.

Packaging is an essential part of the DPA picture. Unlike isolated converters, in which—at power levels higher than approximately 20W—half-, quarter-, and eighth-brick modules dominate, nonisolated converters have yet to coalesce around a de facto packaging standard. For a long time, it looked as if SIP (single-inline-package) modules that mount perpendicular to the plane of the large board might become the norm among nonisolated POLs. Now, however, it is difficult to make the case that such packaging will dominate. Still, betting against the emergence of some kind of de facto packaging standard appears unwise.

Among the factors that are entering the packaging picture is the trend toward lead-free soldering. Ericsson was the first major power-module supplier to offer lead-free dc/dc converters. Lead-free solder melts at higher temperatures than do conventional tin-lead solders. In addition, the range of melting points that lead-free soldering processes can use is smaller than the corresponding range for tin-lead solders. If you use a lead-free process to attach small power modules to a large board, the solder within the modules is likely to reflow as you solder the modules in place. Thermally shielding the small modules can mitigate problems that would otherwise result from this reflow, however. Eventually, all board assemblers will understand how to use lead-free processes to perform feats that are currently uncommon with such processes—for example, soldering small SIP modules to large boards.

Whether MCMs (multichip modules) will enjoy more success in the power-module arena than they have enjoyed elsewhere is another subject of lively debate. Three companies, International Rectifier, Philips, and Power-One offer POLs fabricated with MCM technology. Vicor offers several products, including the PRMs and VTMs of its FPA family as well as a family of BCs in the very small VIC format. Vicor won't say whether it based VIC on MCM technology (Picture).

Power-One points to the fact that its postage-stamp-sized MaXyz MCM POLs fit neatly on the underside of large boards, where board designers can position them immediately beneath the ICs they power, reducing to an absolute minimum the lead length between the dc/dc converter and its load. Because Vicor's VICs connect to the host board via solder bumps on a thin section around the device periphery, you can make a rectangular cutout in the host board to accommodate the thick part of the VIC. This "in-board" mounting technique can limit the thickness of the loaded host board to just 6 mm, which is the thickness of the VIC package itself.

If you decide to design your own POLs, you need to contemplate several interesting packaging alternatives. The most basic question is whether you should mount the components that make up the dc/dc converter directly on the host board or create a subassembly similar to a purchased module. Although using a separate small board or substrate may initially appear to add unnecessary cost, after you consider the differences between the test and burn-in processes for dc/dc converters and large digital boards, you may conclude that the approach saves money.

Author Information

Before he joined EDN, Senior Technical Editor Dan Strassberg, who holds BSEE and MSEE degrees and is a registered Professional Engineer, designed the power subsystems of large automatic-test systems and managed the development of small, linear power supplies for a manufacturer of analog-circuit modules. You can contact Dan via phone at 1-617-558-4205 or e-mail at dstrassberg@edn.com.