Solving power problems for high performance board-level systems

Paul Boughton

The trend towards ever-higher performance in modern semiconductor devices has led to the emergence of a new problem for designers of board-level systems.

A common -- and very simple -- design approach for power distribution in board-level systems is to run either a regulated or a semi-regulated dc voltage across a backplane, providing the input to one or more dc/dc step down converters supplying fully regulated and isolated voltages on each board. On-board power distribution is then done via power planes in the circuit board -- just connect the power pins of your devices to the appropriate power plane. What could be simpler?
The problem with this approach is that it is increasingly inadequate in applications where performance is a major issue. The constant drive for increased performance from silicon vendors has dramatically increased both the clock speeds and level of integration of the devices used in demanding applications. This has led directly to greatly increased power consumption of such devices, coupled with much lower operating voltages, and in extreme cases, as many as eight or more separate voltage rails to track across the PCB. This in its turn has a major impact on the resistive loss through a power distribution plane, since the loss is proportional to the square of the current flow, which has already increased because of both the high power required, and the lower operating voltage. In the simplest cases, the aclassic' isolated converter approach may still work, with a few minor changes to the design. Place the power hungry device close to the converter, and maybe boost the thickness of the metal in the power plane.
Unfortunately, this simple approach soon runs out of steam. A high performance board design may support several high power devices, each requiring multiple tightly regulated low voltage rails. The asimple' solution now becomes rather complex, and in many cases, unworkable. We eventually arrive at a point where tracking fully regulated voltages around a board to several devices each requiring a tightly controlled, low impedance supply will no longer deliver a reliable system solution.

Intermediate Bus Architecture

An alternative approach for such applications is to step down the unregulated supply to a lower level but still higher than required, and bus this intermediate voltage around the board. Additional converters placed close to the apoint of load' of high power devices are used to provide a second stage of (local) regulation. With reduced current flow, voltage drop in the intermediate bus is now reduced, and in any case is much less important since the second stage regulation will now take care of any minor variations. Implementing this IBA (Intermediate Bus Architecture) approach appears quite simple. For example, the board designer could use a single rail12Vdc converter to drive the intermediate bus, and then select whatever lower voltage converters are needed to deliver the second stage conversion point of load voltages from 5Vdc down to 1.2 and even 0.8Vdc. This IBA approach should almost always provide a satisfactory solution to the problems described above.
Unfortunately, such a simple implementation can generate a number of new problems: most notably the significantly increased power losses and the resultant heat dissipation problems that are introduced by the additional stage of regulated power conversion.
Conventional dc/dc converters provide both regulation and isolation, and typically operate with a power efficiency of around 85--87percent. Using such a device at each stage would result in a combined efficiency of 75percent or less (87percent x 87percent). In applications where power losses already present a thermal management burden, this further source of waste heat may not be acceptable.
For an IBA design to achieve efficiencies closer to that of other system architectures, the component dc/dc converters themselves need to be designed to meet the specific requirements of this architecture. By using a step-down dc/dc converter that is isolated but -- crucially -- unregulated as a first stage unit, a great improvement in efficiency can be achieved, since most of the power losses of a dc/dc come from regulation. Such a device effectively performs the functions of a dc transformer, isolating and stepping down the voltage by a fixed ratio.
The efficiency gains which can be achieved by the use of unregulated bus converters, such as Astec's ALQ25, are considerable, especially when coupled with suitable point of load (POL) devices designed specifically for an IBA. This approach has a number of other advantages too. These POL devices provide local regulation but are non-isolated. Such dedicated devices are therefore much smaller and considerably cheaper than a conventional isolated converter. Also, the heat dissipation is spread around the PCB rather than being concentrated in one area, as is common with the isolated converter approach.
Better still, their efficiencies can be much higher, up to 96percent. Comparable efficiencies can be achieved by the IBA converter too, resulting in combined (2-stage) system efficiencies approaching 90percent -- comparable with the best of conventional single stage converters.
As a further advantage, in anything but the most basic system, a properly designed IBA power system can also deliver significant savings in both board space and component cost, where multiple power rails are required. Compared with a traditional architecture, where each rail is supplied from an isolated single or dual rail dc/dc brick converter, the IBA greatly reduces the requirement for isolation. The traditional approach requires more dc/dc converters to have a direct connection to the (unregulated) primary side, with a significantly greater quantity of board space that must be set aside to provide the isolation between primary and secondary sides.
The only drawback to this IBA design approach is a lower tolerance to voltage variations on the input side from the unregulated bus. Variations in this supply voltage, on the primary side, will be translated directly into variations in the intermediate bus voltage, and could potentially compromise the input range limits of the bus converter if too great.

Choosing the IB voltage

Within these constraints, the designer must still make a judgment as to what optimal voltage level should be chosen for the intermediate bus (IB). One obvious consideration is that of I2R losses, and the use of a higher voltage and lower current bus is often preferred, to help minimise these losses.
In practice, a huge range of off-the-shelf POL devices are available to support a 12Vdc IB, somewhat constraining the choice for designers wishing to leverage this extensive base of low cost, high volume parts. However, for load devices operating at voltages as low as 1.2V, the inherent inefficiency of converting from 12Vdc down to this much lower level might make the use of lower IB voltage levels, such as 5Vdc or less, a better choice.
The choice of IB voltage will also tend to optimise the efficiency at either the front end dc/dc converters or at the POL converters, with repercussions for the overall power conversion efficiency across the board. The bus converter -- which provides the IB voltage -- will operate more efficiently when converting to a higher value of IB voltage. However, the decision to use a lower value of IB will improve the power conversion efficiency of the POL devices when converting from a lower input voltage. To arrive at an optimal value of IB voltage, the designer would have to balance both sides of this equation.
The static I2R losses are not the only issues that the designer must consider when designing distributed power architectures. There are also issues that arise from the inability of the circuit to respond dynamically to changing load conditions. Many of the larger devices used in telecommunications, for instance, now feature quite sophisticated apower down' or apower save' modes.
These modes present an intermittent and often quite demanding requirement for a load device to awaken and revert to normal (possibly quite high performance) operation. This in its turn may present a very high dI/dt to the POL converter. If the device in question is located a significant distance from its power source, it can be difficult to guarantee that the inductance presented on the power rail to the device will be sufficiently low to allow the dynamic requirements of the instantaneous load to be supplied. By using a number of small, cheap POL devices positioned close to the most power hungry devices, this effect can be minimised.

Power stability

System developers using any 2-stage conversion design must also be wary of potential instability problems, particularly in designs where a regulated bus converter is used (unlike those just discussed). The load seen by the bus converter driving an IB can be quite dynamic, as this single device is usually driving several active POL regulators. If there is a feedback loop from the POL output back to the IB, it will be quite possible for an overshoot to occur in the IB voltage when a load step is applied by one or more of the POL devices. However, the use of an unregulated bus converter, such as Astec's ALQ25, effectively sidesteps such stability concerns.
Although designers must be wary of potential problems with stability and efficiency, and consider carefully whether the approach suits their application, the IBA can offer significant benefits of power efficiency, space, cost and dynamic performance.

Frank Vondenhoff is European Marketing Manager for Astec Power in The Netherlands. " target="_blank">


Recent Issues