Hacker News Reader

Comment by jmwilson | Hacker News Reader

I wrote an article last month about testing the received wisdom about decoupling capacitors: https://jmw.name/projects/exploring-pdns/

A lot of design guidelines and advice boil down to cargo-cult rules like "sprinkle 100 nF caps everywhere," and many people don't have don't have the tools to measure PDNs and decide if that rule is actually good enough.

I made a custom PCB to emulate different configurations of capacitors of various types and layouts. With a spectrum analyzer and tracking generator, it's possible to measure and visualize several points that are important: what does "physically close" mean, the effect of DC bias on capacitors, and how does parallelizing capacitors together affect the circuit. The author of this is using an oscilloscope to look at time domain behavior, but generally I think of these things in the frequency domain.

View on HN

Thanks James, nice article. (I had not seen the TLA "PDN" before but now I know you mean "Power Distribution Network".)

VNAs have become remarkably cheap; here's one that < $ 60 https://www.amazon.com/50KHz-900MHz-Analyzer-Measuring-Param... I have one of these, it works very well.

Fancier ones are of course available. This one goes to 6.3 GHz https://www.amazon.com/LiteVNA-64-Analyzer-50KHz-6-3GHz-Port... and is $ 180.

I have this 'fancy' one: https://www.tindie.com/products/hcxqsgroup/nanorfe-vna6000/

One trick I learned from the RF guys is: put two same value capacitors in parallel; it doubles the capacitance, and also reduces by about half the ESL (inductors in parallel rule). Clever.

Very cool and thanks for sharing. From practice and hearsay I had reason to expect that resonance at ~900MHz from standard FR4 stackup, but seeing it on the graph really clarifies.

Nice article! For the decade test, did you try just four 10uF? I've seen others perform similar tests where four equal values were from at least as good to much better.

I have data collected for 3x MLCC 10 uF + 1x 100 nF but need to make the charts and update the article. Short answer is it's basically just as good and has fewer resonance peaks. I'd also like to do the test with some tantalum 10 uFs to throw in some ESR and see if that tames the impedance spikes even more.

I'm experiencing minor flashbacks to the EE intro class where they made us calculate RC circuits. Objectively I know a .1µF capacitor doesn't affect circuit voltage that much but my inner 19 year old is screaming.

Your article says "VNAs are very expensive." Does this mean something like the NanoVNA [1] is inadequate for this measurement?

[1] https://nanovna.com

I do mention the NanoVNA as a measurement option that has better accuracy since it can do phase measurement. It's an amazing tool, though its lower limit for frequency is 50 kHz and I'm not sure how its dynamic range holds up there. Ideally you want the option to go very low for this measurement. Like most VNAs sold today, the NanoVNA is really designed for looking at RF and microwave circuits.

I only know two commercial VNAs in-production that are made specifically for this kind of work: the Keysight E5061B (5 Hz - 3 GHz, $50000+ when optioned out appropriately) and the Omicron Bode 100 (1 Hz - 50 MHz, I think $5k-10k).

I assume you don't actually need all of the transmission line s-parameters at low frequency. In which case, you can use a frequency response analyzer (FRA), of which there are a few in production. The gain and phase measurements are enough to plot transfer functions, impedance, Bode, and Nyquist.

The FRAs span the same price range as the VNAs you listed. Having a low-cost version of each can extend the frequency range over which you can make good measurements compared to just one tool of similar total price.

The Bode 100 sacrifices some low and high frequency range and performance to be a 2-in-1.

What sacrifice would you say that the Bode 100 makes in performance (Especially compared to the E5061B)?

All test equipment has reduced performance near its limits. I probably should have just said range because it's implied.

I think the Omicron Bode 100 is the best bang for buck for a proper VNA these days.

Id love to see a practical test of how we'll certain IC's work with different capacitor combinations. Because in the end, the IC operation is the important part.

Reply on news.ycombinator.com