When designing the power portion of the board in your embedded systemdesign, it is tempting to simply select standard modules from acatalog. These days, many suppliers offer a wide range of excellentproducts, whether isolated bricks or non-isolated point-of-load modules(POLs).
Efficiencies are high, quality and reliability are good, costs arecompetitive ” what more could you wish for? But as always, things arenot as simple as they may seem. Although a power system using standardmodules can be a good solution in many cases, a custom design canusually offer better performance at lower cost. This article examinessome of the issues involved in making an informed trade-off betweenusing a standard or custom solution.
For this discussion we will consider a typical card power system in aproduct using distributed power architecture, where each card operatesfrom an input voltage of -48V. This power architecture is very commonin telecom and high-end computing applications, and is specified in theAdvanced Telecom Compute Architecture (ATCA) equipment standard.
In most applications, each card used in a distributed power systemmust provide electrical isolation between the -48V input and the lowvoltage outputs to meet safety requirements. The basic requirements forthis example are listed in Table 1,below.
|Table1 ” Card power system requirements|
The input requirements are set by the application and are generallycommon to each card in the system. The other requirements are differentfor each card, to suit the components used.
Standard Module Solution
Figure 1 below shows a possiblesolution to meet the power requirements in Table 1 above using standardpower modules. In this example, two bricks are used for the high poweroutputs with POL modules for the other three outputs. This allows highefficiency, and can easily meet the requirement for 2.5V to startbefore 3.3V.
|Figure1 ” On-card power system using standard modules|
Note that the brick power outputs must be sufficient to drive thePOLs as well as the direct outputs. Assuming 90% efficiency for thePOLs, the 2.5V brick output must be at least 19A (17A direct plus 2Afor the 1.5V POL). Similarly, the 3.3V brick must be at least 17A (11Adirect plus 5.7A for the 1.2V and 1.8V POLs).
This design approach offers the advantage of minimum design time,and a wide range of bricks and POLs are available from multiplesuppliers. If required, the output voltage or current can be easilychanged during development by using a different module. However, thebricks and POLs function independently during startup and faults.
External circuits are needed to meet system specs for input startupand shutdown, inrush current limiting, output sequencing andprotection. For noise sensitive applications it can be desirable tosynchronize the power circuit switching frequency to a system clock,but this cannot be done with standard bricks or POLs. Each module setsits own frequency, and each supplier is different.
Since bricks and POLs are only available in specific power levels,it is usually necessary to choose a module with somewhat higher outputpower than the card requires. It is also necessary to properly accountfor the thermal derating required in the application, which may bedifferent for each supplier and which complicates qualification ofmultiple sources. Because of this, and because of the external circuitsneeded, a power system using standard modules is usually more expensivethan a custom solution.
Custom Power Solution
As an alternative to standard modules, a custom power solution offersthe possibility of optimizing the design to exactly meet the needs ofthe card. Figure 2 below showsa custom power module to meet the same card power requirements (seeTable 1). The power module includes all the DC-DC conversion andisolation functions as well as the sequencing and protection, replacingthe bricks and POLs used in Figure 1 previously.
|Figure2 – Custom on-card power module|
Note that the input functions in this example are not includedwithin the custom power module. This allows the EMI filtering to beeasily optimized during system integration testing without the need toredesign the power module, and is a good approach for noise sensitiveapplications.
In other applications the noise filtering may be less critical andthe power module can include the input functions, as shown in Figure 3, below . Even in this case,the input fuses are normally external to the module, since they must belocated as close as possible to the card input connector to meet safetyrequirements.
|Figure3 ” Fully custom on-card power solution|
The custom design approach avoids duplication of functionality andallows the power circuits to be optimized to meet the card requirementsat minimum cost. Output sequencing and fault protection can beguaranteed by the module design, and switching noise can be reduced bydriving all power conversion stages from a single clock.
Furthermore, a custom power module requires less PCB area thanstandard modules, and the physical packaging ” dimensions, pin-out,mounting, cooling – can be optimized to suit the application. However,significant design time and effort is required, so that this approachis better suited for medium to high volume applications.
Discrete Power Solution
Another option is to mount the power components directly on the card,rather than using a separate power module. The design can be optimizedto exactly meet the needs of the card, and the material cost can bereduced by eliminating the power module PCB.
As shown in Figure 4 below ,the block diagram is essentially the same as for the full custom modulein Figure 3. However, the components used to implement the functionsmay be significantly different because they must be compatible with theoverall card assembly process.
|Figure4 ” Discrete power circuit|
For example, it is usually not possible to embed a planartransformer or power inductor within the PCB, as is commonly done forpower modules. The number of layers and the copper thickness of the PCBmay also make it less useful as a heat sink for the powersemiconductors.
The discrete power design approach also makes testing andtroubleshooting more difficult, since the outputs are directlyconnected to the rest of the card. Functions such as overvoltageprotection or current limiting cannot easily be tested without riskingdamage to the card in case of a fault, and even output ripple and loadregulation are not easy to measure. If there is a fault,troubleshooting must consider the possibility that other circuits onthe card may have been damaged. In general, this type of power systemdesign is best suited to high volume, low cost boards where testing isless extensive, and troubleshooting and repair are not necessary.
|Table2 ” Comparison of standard, custom and discrete power solutions|
As shown in Table 2 above ,which summarizies the main advantages (shown in bold) and disadvantagesof each design approach, while the best choice of a power systemdepends on a wide range of factors, some general conclusions are clear.
A solution based on standard modules offers good flexibility and theshortest design time, but can be expensive. At the other extreme, adiscrete solution offers the lowest cost but is more difficult todesign, manufacture and test. In many applications a custom powermodule can offer a good balance, with excellent performance and ease ofmanufacture at a lower cost than standard modules.
For a PDF version of this article, go to Onboardpower: custom or standard?
David Cooper is ApplicationsManager at Saft Power Systems.