System Powering Architectures Part V: DC-DC Conversion

The selection matrix for on-board conversion choices is quite impressive. The product portfolio has been expanding as new technologies are introduced and revised concepts are deployed. Developers unfamiliar with a product line or its history can be overwhelmed by the available component choices for their applications. The utilization of application engineers or a comprehensive online Web-based guideline is almost mandatory to make component selection simpler and less time-consuming for the developer.

Early generation products were designed for the telecommunications industry where board height has been the constraining form factor. Height limitations of 0.5 in. for domestic telecom and 0.38 in. for the Japanese market define telecom platform requirements. Either thermally encapsulated, to minimize hot-spot generation, or open-frame constructions are available (Fig. 1). The encapsulated product is ideal for applications where externally adaptable heat sinking can appreciably extend the power capacity of the converter. It's also well suited for outdoor environments and mounting to cooling baseplates. The open-frame construction is a second-generation product, geared to minimize airflow blockage around the converter. This is best suited in applications that provide adequate airflow to cool the power components of the power module. The open-frame construction is lower cost, lighter, and requires less height.

Third-generation converters have application designs that are flexible in height requirements where p. c. board real estate is at a premium. The Straight In-Line Packages (SIP) converter was designed especially for this application.

Each of these products has unique characteristics, and the usage choice is highly dependent on the constraints of the application. Therefore, instead of recommending one type of product over another, it's best to follow some general guidelines:

Design the on-board dc-dc converter into the system's fault group structure. Treat this conversion as just another component of the failure group. It's likely that this wouldn't require redundant conversion, saving expense, complexity, and real estate. In most applications, the failure group will be the specific circuit pack. But in many cases a number of circuit packs perform as a single failure group. In those instances, there may not be a need to provide conversion separately on each board.

Don't push the converter's power consumption capability. Obtain the vendor's thermal recommendations for the device, ensuring reasonable thermal stresses are followed. Caveat: MTBF is halved for the rise of every 10°C.

Powering Fast Transient Circuits

The latest generation of on-board converters is designed for powering high power processors, DSPs, and other components that have very rapid transition states. The major issues are performance, cost, and size.

Processor voltage requirements are decreasing and substantial increases in current consumption are complementing this trend. Some predict trends will continue for the next few years. Future processors will continue to consume current levels an order of magnitude higher than today's devices, and will dissipate much higher power levels because the voltage rail requirements will not decrease proportionately. Soon we will see devices that will require substantially better performance than are required today.

Component families are divided between two platforms. SIPs are used in applications where sufficient height is available. To increase power capacity, many SIPs are parallelable — offering a solution that provides high-current capacity in a space-efficient package (Fig. 2).

Although SIPs provide excellent performance, their transient response capability is limited by the interconnect inductance between the converter and the circuit board placed mating connector. As current consumption and rapid load transitions increase, a better solution will be required that isn't limited by the interconnect inductance of the connectorized SIP converters. The flat surface-mount converter is one obvious choice for these applications because it's soldered directly onto the processor board, minimizing interconnect inductance. However, this performance improvement has its tradeoffs. The flat, surface-mount converter requires substantially more real estate and heavy interconnect copper paths for parallelable objectives.

Custom design is also an alternative. Prudent placement of the output conversion stage near the mating surface, short interconnect leads, and soldering of the leads on both circuit boards dramatically improved the transient performance by reducing parasitic inductance.

The SIP feature significantly reduces the motherboard's space requirement. In this application, proximity of the converter to the processor was limited by the placement of more critical components nearby. Computing speed is proportional to path length and thus ASICs, flash, and other high-speed components must be positioned near the processor to achieve optimized performance. Although the converter also needs to be close for the required performance improvements, it's seldom the selected choice near the processor. Therefore, performance is limited by how close the converter could be located in the specific application.

Topology Choices

Device and Power Supply manufacturers offer a variety of choices for powering processors, spanning from providing reference circuits for users electing to place the designs on the motherboards to complete power supply packages for 3.3V to 12V inputs. Although the circuit concepts may be different, employing a single non-isolated conversion stage is a common practice. Some vendors also provide parallelability to increase available current capacity, as shown in Fig. 3. Seek transient and stability modeling from the power vendor to quantify and ensure desired converter performance.

As distribution voltages increase to 48V, the number of choices rapidly diminishes. At 48V, an isolated topology is necessary because of the higher voltage transfer ratio. Buck type circuits are impractical in these situations. Additionally, many new applications require high availability and a solution that accommodates dual 48Vdc input feeders. The block diagram in Fig. 4 is one concept.

The concept is a micro-level distributed on-board architecture. Diode OR-ing establishes true separation between the two input busses and holds-up the input to the converter powering the processor in case of an upstream fault.

Fast Transient Response

The most obvious, yet still seldom used, solution to cope with fast transient requirements is to add external capacitors as close to the load as possible. The typical value can be thousands or even tens of thousands of μF. These capacitors can be either tantalum or a combination of ceramic and electrolytic types. A recent worldwide shortage of tantalum capacitors forces developers to choose the ceramic and electrolytic approach for procurement reasons. Although these external capacitors provide an effective remedy to the fast transient response, they have drawbacks. First, large capacitance directly translates to occupying more real estate on the motherboard, a major premium in modern electronic systems. Second, more capacitance with low ESR usually leads to higher cost. Last, adding large amounts of external capacitance may alter the dynamic characteristics of the converter, which could cause a stability problem.

Some efforts have improved the transient response by using the current information of the converter's output section. A problem with this is the load current changes its value so rapidly that the control circuit can't respond fast enough. Therefore, if the information of the fast change load current can be expedited to the controller, the converter will respond much faster, reducing the need for external storage. One possible implementation is to introduce a current feed-forward signal. Another method is to feed-forward the output capacitor current.

The above-mentioned methods offer various ways to solve the fast transient response issue. In a final system design, tradeoffs, such as cost, space, circuit complexity, etc., should be taken into account. For some applications, one of the solutions may be adequate; for others, designers may have to use an alternative method or even a combination of several methods, depending on the circumstance.

Stability

System stability is another important issue for the application of power supplies, especially for dc-dc power modules. Sometimes users find that a dc-dc converter module meets every technical specification and performs well during the evaluation stage. As the design approaches completion, more components are added into the product, and more tradeoffs, such as noise, transient, voltage ripple, etc., occur. When the system is integrated, the complete circuit may oscillate or show relatively low-phase or gain margins.

A power module is not an ideal voltage source. Any components connected to the output of the converter will interact with the module. This interaction inevitably alters the dynamic characteristics, like stability. Fig. 5 shows the Bode plot of a dc-dc converter module with and without the external capacitance. The module by itself has a crossover frequency about 3 kHz and the phase margin about 92°. An addition of 10,000 μF external capacitance changes the Bode plot drastically.

In Fig. 6, the crossover frequency drops down to 1.3 kHz, and the phase margin to about 30°. Such a change, if not properly accounted for, could cause a serious design issue.

It's important to have a tool for a power module to predict the overall performance of the final system without the system being physically built. To predict system performance, users must supply the model's load impedance characteristics in frequency domain. Plotting the impedance characteristics on top of the Young Stability Curve (YSC) immediately gives the crossover frequency and stability margins. This method works for resistive, capacitive, and/or inductive loads.

Standard Based Custom Converters

Challenges increase with greater power consumption demands, multiple output voltage requirements, and limited p. c. board space for power conversion. In many applications, insufficient space on the board requires a custom solution. The critical issues here are expense reduction, minimizing development risk, and timely product availability.

To satisfy these concerns, many designers rely on the availability of standard circuits and packaging ideas. A building block concept is deployed where a number of standard circuits are combined to achieve the desired function. Protection and monitoring features are also derivatives of standardized circuit concepts, customized to meet specific requirements.

A developer and vendor partnership offers the lowest risk approach for product development. With defined constraints and a mutual agreement, the development of a package that is best for the required component layout is possible. Fig. 7, on page 62, shows examples of such processes and products using the building block concepts. The functional diagram shows how a standard platform, in this case an on-board converter circuit, combined with a combination of standard circuits and well-understood post regulation concepts derived two separate products.

The building block concept is readily seen in the symmetrical layout. The converter is hot pluggable, current-shared and diode isolated on each output for N+1 redundancy. Alarm circuitry monitors each output and warns the system if a failure occurs. Visual indicators on the front faceplate also assist the craftsperson. Normally, the developer and vendor jointly define these products. Vendor participation ensures feasibility and reduces the risk associated with custom product development.

In other applications, the custom product may be designed using available standard on-board modules. These are favored because of significantly lower risk, much faster time-to-market, and development cost savings. The example in Fig. 8 is a power supply for a radio amplifier. The power supply is attached to the heat sink enclosure of the outdoor amplifier assembly. Metal encased modules were the ideal solution for the power converters since they are directly mountable to the amplifier heat sink. The circuit board enclosure provided external circuitry for EMI filtering, remote control, paralleling of the two modules for greater power capacity and connectorization to the external environment.

For more information on this article, CIRCLE 335 on Reader Service Card