I recently ran into an old friend who is an experienced analog engineer who designs power subsystems for high-reliability servers. He said that a typical rack-mount board he designs with a width of 19 inches draws 100A. I was not surprised. We all know that today's processors are power hungry because there are millions of active devices on the chip, and the power consumption adds up. But my friend disclosed another statistic: a typical board he designs has about 30 separate power networks. Each power network has a different nominal supply voltage, accuracy, and regulation; in some cases, these nominal voltages differ by only a few tenths of a volt. Furthermore, each power network requires its own regulator and a series of decoupling capacitors to control the bypass impedance within a bandwidth from nearly DC to hundreds of kilohertz. Designers must analyze and implement the power supply and return paths for each power network, as well as a large number of PCB board traces. In the final design, the traces and capacitors of the DC power subsystem occupy a large part of the board area. Designers must carefully model all of these factors to ensure that the current path is appropriate and the IR voltage drop is small. This is not an easy task when reaching these current levels. Yet there is a conundrum between a high-quality power subsystem and its distribution system. Although power delivery is an integral function in any system, it is not directly appreciated or recognized by the user. Users demand additional features, functions, and performance; power delivery is viewed as an inherent part of the design. Adding features is good for marketing and generates more profit, but the component cost and board area of the power network do not have these benefits. In fact, some would view the board area occupied by the power subsystem as a meaningless burden, like the accounting department or the mail room. I hope that you as a system designer or circuit designer can have a significant impact on the selection of components on the bill of materials. My friend pointed out that there are a few basic things you can do to minimize the burden on the power network. First, help the power subsystem designer develop a set of basic regulators (either linear or switching) so that you can reuse these designs on the board. To make this work worthwhile, you should also balance the current loads for each nominal voltage to be in the same range, because you will not find an economical design that can support both 10mA and 1A loads. Second, you must discipline yourself. In the vast world of ICs, avoid greed. Choose components that share fewer common nominal voltages. Strive to find parts that meet data sheet specifications, even if they have a 5% line regulation, because tighter specifications limit the flexibility of the power supply. Use voltage combinations of standard multi-output supplies (preferably from multiple manufacturers) so that different output currents and regulations can be selected in pairs to support multiple ICs, not just one or two. You have to make trade-offs in two areas: on the one hand, you want to choose the most suitable IC, and on the other hand, you want to get the features and functions you need after understanding the impact of its unique power requirements on the power network design. In other words, you have to make the trade-offs that engineers often make (these trade-offs are sometimes obvious at a glance, but they are usually difficult to determine): weigh the pros and cons, make judgments on the trade-offs, and strive to achieve the best point of power, price, and performance that meets market requirements.
|