Below are the questions asked during the live event, along with their respective answers.
Q: Is the best power supply the one with the lowest impedance?
A: Low impedance does reduce noise ripple, however, the power supply is only a part of the PI Ecosystem and it is important to remember the parallel L with C resonance that can occur at frequencies beyond the control loop bandwidth fo the power supply. Lowering the impedance of the supply means that more bulk capacitance will be needed to reach the low impedance of the power supply and avoid a resonant peak with the active L of the supply. It is better to design the VRM for the required impedance of the load so that the cost of the bulk capacitors and decoupling can be minimized.
Q: If the data sheet doesn’t provide the active inductance of the VRM output, then how does a designer find this value for calculating the bulk capacitor for flat impedance?
A: The output active inductance of the power supply may not be in the datasheet, but it is easy to measure. Measurements are also a great way of verifying that a component is working as expected. To measure the active output inductance, one can build a small engineering characterization board during the pre-layout stage, or measure directly on an existing post-fabrication board. The easiest place to measure is at the pads of the bulk capacitor. The impedance is measured using the 2-port shunt method for low impedances on a network analyzer like the E5061B. Measuring the impedance in the off state shows the downward slope of the bulk capacitance at low frequencies. Measuring the impedance with the power supply on will show a small peak where the VRM active output inductance is resonating with the parallel bulk C capacitance. If the peak is too small, then some of the bulk capacitance can be removed until the rising edge of the inductive side of the peak is large enough to measure an estimate of the VRM’s active output inductance. This measured inductance L can then be used for the simple VRM R series L model to estimate the PDN capacitors needed for a flat impedance design.
Q: Does it matter where the decoupling capacitors are placed?
A: I often hear that it is low frequency, so it does not matter where the capacitors are placed. However, it depends, and as margins decrease even small parasitics can start to impact performance. Noise is lost energy, so a correctly designed system should deliver signals from source to load or in the SI world from Tx to Rx with the least amount of noise. The best place to stop noise is at the source so looking at the PDN, the two noise main noise sources are the VRM and the Sink. Place the bulk capacitors near the load to provide a stable power supply load and take care of switching ripple. Place higher frequency decoupling next to the sink with the smallest de-caps the closest to take advantage of the lowest inductance path.
Q: If oscillations on the power rails are a source of noise/rogue waves do you know of any standards being developed to address this?
A: EMC/EMI standards like CISPR 25 for conducted radiated emissions may not check for all possible operating modes, so they typically only show a nominal operation pass/fail. The aerospace/defense industry does have some Non-Invasive Stability tests that are used for verifying that the power supply is sufficiently stable with enough phase margin. Chip vendors are starting to specify target impedance masks, but these are not standards yet. I believe that PI is about 10 years behind the SI world in terms of using the frequency domain type S-parameter data for design/test. Signal Integrity standards are making use of channel S-parameter data, and I expect to see the same happening with standards for Power Integrity to prevent unwanted resonances and rogue wave situations.
Q: For applications where materials are specified for EMI Shielding, are there rules of thumb for translating between voltage noise limitations and system-level shielding (or shielding effectiveness) requirements?
A: I am not an expert on EMI shielding, but if designs did a better job of ensuring that the required power was delivered from source to sink and not radiated or causing resonances then the amount of shielding could be greatly reduced. Shielding adds cost, and should not be used as a band-aid to fix a poor design.