Measuring Droop

Rajen Murugan of Dell writes:

What is the best technique to make dc-voltage measurement on a power rail? Will a four-point measurement technique be the most accurate?

You can simply apply a voltmeter across the power and ground terminals of any bypass capacitor in the area you wish to measure. To measure the power to a chip, apply the voltmeter to the actual vias that touch your chip's power and ground balls or pins. The difference between the measured voltage at the output terminals of your power supply and that at the input terminals of your CPU is called the droop in your power system. Values as large as several tens or even hundreds of millivolts are common.

picture courtesy EDN MagazineThe dc source impedance of the power system lies far below 1 Ω, so you do not need a fancy high-input-impedance voltmeter. Any meter with lots of digits of precision will do.

Before making extremely fine comparative measurements, experiment with your voltmeter to see how temperature and proximity to the emissions from your system affect it. Keep the instrument in a well-protected, consistent environment. Long leads running to the voltmeter work fine for ordinary dc measurements. Practically no current runs down the lead wires so their resistance is negligible. Your board performs its worst when hot.

With your leads touching power and ground terminals, every measurement you make includes a fixed dc offset equal to the power-supply voltage. When tracking down millivolts of power-supply droop, that dc offset obfuscates your measurements. For example, using a three-digit meter to read voltages on a 2.5-V rail, the smallest observable value of droop equals the minimum resolution of the meter, 0.01 V. That value is not good enough.

Why not more directly measure droop? With one terminal of your meter touching the output terminal of the power supply and the other at the power entrance to your CPU, your meter shows droop in the power rail without the annoying dc offset. Now turn up the gain on your three-digit meter to show much smaller observable changes. Don't forget to also measure droop in the ground connection from the ground terminal of the power supply to the ground on the CPU, and add that to the power-rail-droop number to compute total droop.

Here's how you can measure a bare pc board. This technique requires that you know the worst-case power-supply current that each chip in your system draws.

First, short from power to ground at the power-supply terminals. Next, connect a regulated power source to Chip A. Connect to every power via and every ground via in that chip's footprint. Place an ammeter in series with the power source, and adjust its current limit to the exact current that the target chip draws. Source current now circulates backward from the chip through the power system, the exact opposite of what happens in a real system, but the voltage drops will perfectly mimic what happens in real life.

You can now observe realistic droop voltages produced at any location due to the current that Chip A draws. This four-terminal technique removes all dc bias from your measurement, so you obtain full meter resolution. Repeat the measurements for current injected at other chip locations. Superimposing all the results produces a perfect map of total droop in the finished system.