Power integrity measurement objects and measurement contents

PI (Power Integrity) used to be a topic of signal integrity analysis, but because PI is complex and critical enough, it has now been taken out as a separate topic for research. Rapid and accurate simulation of power integrity is still a difficult problem to be solved.

For high-speed digital circuits and systems, PI's research object is the power distribution network PDN (Power Distribution Network). Taking a laptop as an example, the AC to DC power adapter supplies a DC power supply of about 16V to the computer motherboard. The power distribution network on the motherboard must convert this 16V DC power supply into DC power supplies of various voltages (such as: +-5V, +1.5V, +1.8V, +1.2V, etc.) to power the CPU and various chips. The CPU and IC consume a lot of power, and the power consumption is dynamic. The instantaneous current may be large or small, but the voltage must be stable (that is, the ripple and noise must be small) to keep the CPU and IC working properly. This puts stringent requirements on the PDN.

To measure the performance of a PDN, you first need to use an oscilloscope to test the power ripple and noise of the CPU and IC pins. However, to accurately measure the performance of a PDN, you also need to test the output impedance of the PDN (impedance that varies with frequency) and the transmission impedance of the PDN (impedance that also varies with frequency), just like characterizing a single-port network or a two-port network. Since most of today's PDNs are switching power supply structures, you also need to measure the loop gain of the PDN or key DC-to-DC converter devices.

To summarize, the measurement object of power integrity is the power distribution network PDN. The main measurement content includes four parts:

● Ripple and noise measurement;

● Measurement of output impedance;

● Loop gain measurement;

● Measurement of performance parameters of filter components (capacitors/ferrite beads, etc.).

Power Supply Ripple and Noise Measurements

Power ripple and power noise are concepts that are easily confused. As shown in Figure 2 below, the blue waveform is the power ripple and the red waveform is the power noise. The frequency of the power ripple is the fundamental and harmonic wave of the switching frequency, while the frequency component of the noise is higher than the ripple. It is caused by many factors, such as transient current generated by the switching of the high-speed I/O of the chip on the board, the parasitic inductance of the power supply network, and the electromagnetic radiation between the power plane and the ground plane. Therefore, the power output measured on the PMU side is ripple, while the power noise is measured on the SINK end (the end of the power-consuming chip, such as AP, EMMC, MODEM, etc.).

Today's electronic circuits (such as mobile phones, servers, etc.) have higher switching speeds and signal slew rates than before, while chip packaging and signal swings are getting smaller and smaller, making them more sensitive to noise. Therefore, today's circuit designers are more concerned about the impact of power supply noise than before. Real-time oscilloscopes are a common tool for measuring power supply noise, but if used incorrectly, they may lead to completely wrong measurement results.

Since the power supply noise bandwidth is very wide, many people will choose an oscilloscope to measure the power supply noise. However, it cannot be ignored that the real-time broadband digital oscilloscope and its probe have their own inherent noise. If the noise to be measured is of the same order of magnitude as the noise of the oscilloscope and probe, it will be very difficult to make an accurate measurement.

The main noise of an oscilloscope comes from two aspects: the noise of the oscilloscope itself and the noise of the probe.

All real-time oscilloscopes use attenuators and amplifiers to adjust the vertical range. After the attenuation is set, the noise of the oscilloscope itself will be amplified. For example, when the attenuator is not used, the basic range of the oscilloscope is 5mV/division. Assume that the bottom noise of the oscilloscope at this time is 500uVRMS. When the range is changed to 50mV/division, the oscilloscope will add a 10:1 attenuator in the input circuit. In order to display the correct voltage signal, the oscilloscope will amplify the signal 10 times when it is finally displayed. Therefore, the bottom noise of the oscilloscope now looks like 5mVRMS. Therefore, when measuring noise, the most sensitive range of the oscilloscope should be used as much as possible. However, the oscilloscope usually does not have enough offset range at the most sensitive range to pull the DC voltage to the center of the oscilloscope screen for testing, so it is usually necessary to use the AC coupling function of the oscilloscope to filter out the DC level and only measure the AC component.

There are 12-bit oscilloscopes on the market now, such as the Agilent 9000H series oscilloscope. Its noise is relatively small, only 0.7v@100mv/grid, so a 12-bit oscilloscope is the best choice.

For the same reason, you should also use a 1:1 probe instead of the 10:1 probe that comes standard with the oscilloscope when measuring power supplies. Otherwise, the oscilloscope's noise will also be amplified.

The noise brought by the probe is coupled in front of the attenuator, so no matter how much the attenuation ratio is set, the noise contributed by the probe is certain. However, under some incorrect usage methods, the probe may bring additional noise. A typical example is the use of a long ground wire. In order to facilitate testing, the passive probe of the oscilloscope usually uses a long ground wire in the form of an alligator clip of about 15cm, but this is not applicable to the test of power supply ripple, especially when there is a switching power supply on the board. Since the switching of the switching power supply will generate a large amount of electromagnetic radiation in space, and the long ground wire of the oscilloscope probe is just equivalent to an antenna, it will introduce large electromagnetic interference from space into the measurement circuit. A simple verification method is to connect the ground wire and the front end of the probe together, close to the circuit under test (not directly in contact), you may see a relatively large switching noise on the oscilloscope. Therefore, the shortest possible ground wire should be used during the measurement process.

Now many DUTs require the measurement of ripple and noise with a peak value of several millivolts. For example, some SerDes above 10Gbps require 3mv peak-to-peak power ripple and noise. At this time, it is best to use coaxial cable for measurement. Although the impedance of coaxial cable is only 50 ohms, for the millivolt-level DUT, the load effect is very small and the test accuracy is very high.

However, when using coaxial cable and the oscilloscope is set to 50 ohm input impedance, the oscilloscope is DC coupled. There are two ways to deal with this:

First, place a capacitor at the contact point of the power supply under test. One side of the capacitor is connected to the device under test, and the other side is in contact with the coaxial cable. Generally, a 0.1uF capacitor is sufficient.

Second, make a power test probe. It is best to make a small PCB with SMA connectors at both ends and the middle exposed to place capacitors. Figure 3 is an example of a homemade probe.

The last point to note is that power supply tests usually specify ripple and noise within a certain frequency range, such as within 20 MHz, and the bandwidth of a general oscilloscope is greater than this requirement. Therefore, the bandwidth limit function of the oscilloscope can be turned on during testing, which will also have a better effect on reducing high-frequency noise.

To summarize, for the test of power supply ripple noise, we usually need to pay attention to the following points:

● Try to use homemade power test probes

● Try to use a 12-bit oscilloscope

● Try to use the most sensitive range of the oscilloscope;

● Use AC coupling function as much as possible;

● Try to use a probe with a small attenuation ratio;

● The ground wire of the probe should be as short as possible;

● Use bandwidth limiting function as needed;

Measurement of output impedance of power distribution network (PDN)

To measure the performance of a PDN, it is not enough to just use an oscilloscope to test the power ripple and noise of the CPU and IC pins, and there is no way to locate the problem after it occurs. To accurately measure the performance of a PDN, it is also necessary to test the output impedance of the PDN (impedance that varies with frequency) and the transmission impedance of the PDN (impedance that also varies with frequency), just like characterizing a single-port network or a two-port network to characterize the PDN. This requires the use of a network analyzer tool.

There are two major challenges in testing PDN with a network analyzer:

1. The output impedance and transmission impedance of PDN are in the milliohm level (usually around 2m ohms), so it is difficult to test them accurately.

2. PDN works with DC voltage, that is, with bias, and the network analyzer needs to have the bias measurement function.

When testing the output impedance of milliohm level with a network analyzer, you cannot simply use the one-port test method because the impedance is too small and the reflection is too large. In this case, a better method is to use the two-port test method and use S21 instead of S11 during the test.

Assuming that the inductance of the probe test cable is approximately 0, and Z(DUT) is much smaller than Zo (VNA port impedance), the PDN output impedance is calculated as follows:

ZDUT=Z11=S21x25

The output impedance in the milliohm range is tested using a network analyzer, also using the dual-port test method.

Assuming that the inductance of the probe test cable is approximately 0, Z11, Z21, and Z22 are much smaller than Zo, the calculation formula for the PDN transmission impedance is as follows: