Parallel caps in schematics

We had a long thread all about this recently - where have you been?!

Briefly, real capacitors are not perfect, they have a small associated resistance and inductance (ESR and ESL). By putting a them in parallel, you end up with lower overall resistance and inductance than a single equivalent capacitor, which is important when used for supply bypassing.

A cap is not really a pure C, it is a series Z=ESR+j*(ESL*W-1/WC). Put any three Z in parallel and you end up with Z/3=ESR/3+j*((ESL/3)*W-1/(W*(3C))). Do you see what that buys you?

They are there to bypass noise. The traces on the board are inductors and that reactance causes isolated areas where noise on a trace can be a problem. SOP for logic devices - some designers just stick them in there to be safe.

The ADC you are working with has frequency components from DC to nearly 100 MHz.. The 10 uf capacitor has a self-resonance down in the hertz region, the 0.1 uf in the kilohertz region, and the 470 pf up somewhere around 70-80 MHz. with half-inch leads. The 10 uf is worthless for bypassing at MHz. and the 470 pf is worthless bypassing at low frequencies. Put them in parallel and you have pretty good bypassing from DC to nearly 100 MHz..

As to the voltage, you can't buy picofarad capacitors in voltages less than

Jim

Capacitors have series inductance, so for good high frequency bypassing it helps to put them in parallel. Also the multiple caps can be put in different places, bypassing multiple power pins and such.

There's less justification for using different values; none in my opinion. One big cap, the 10 uF guy, might be helpful if the supply current has serious low-frequency components, unlikely here. The other caps should be physically small and electrically large.

If I were laying out a board for this chip, I'd use a good ground plane and a split power plane with Avdd inside the chip footprint and Dvdd outside, with a bunch of 0.33 uF 0603 caps bypassing each power pour. Or separate power planes, above and below ground, if you can spare the layers.

John

Several identical caps connected in the parallel are making for the worst of the possible combinations of resonances and antiresonances. If a very solid wideband decoupling is required, the paralleled caps should have the values different from each other by 10 times or so.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

No. Putting different caps in parallel just moves the resonances around; Spice it. But a good power-ground plane structure, with large-value bypass caps scattered about, is a very stiff, low-Q structure, and Spice is a poor-to-worthless model of that reality.

I occasionally put SMA footprints on my power/ground planes, and TDR the planes without and with bypass caps, and later measure the actual HF noise on operating boards. Try it, it's educational. Most of the literature on bypassing is absurd.

John

It affects Q as well. Many small humps around could be better then several big humps concentrated in the particular frequency area. The common 0.1uFs resonate at about 10Mhz; just where I don't want it.

My remark is related to the colocated caps for the very wideband decoupling of a part. Scattered caps are the different story.

What book do you dislike most?

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

Because it's always been wrong, or because newer technology has made it obsolete?

Hmmm...

You suggest Spicing, then point out that Spice is a poor tool for analyzing real-worl layout situations.

That just depends on how much effort you put into accurately modeling the trace,lead and component body transmission line qualities.

But it's a *series* resonance, which is not a bad thing. I.e., the decoupling is quite good over a range of frequencies, and excellent at

Johnson's "Black Magic" is especially annoying.

John

Spice will demonstrate that putting different-value caps in parallel doesn't nuke resonances. But isolated capacitors on a schematic are still very different from the physical reality of a multilayer pcb, which Spice can't really model.

Spice is fine when you really understand the lumped components.

John

Exactly. If you could get a 10 farad 0402 ceramic capacitor, and its SRF was in the kilohertz, it would still be a great bypass at 100 MHz.

John

Any specific problems with the book?

This site might help you,heres a qoute from it.

"RF practice (and serious microprocessor practice in some cases) is to group several capacitors of different values together to combine the characteristics of each = effectively a rather broad bandpass filter. Use of small ceramics and larger valued distributed electrolytics (tantalum for the brave, solid aluminium for the wise, wet electrolytic for the adventurous) can be useful."

Heres the site.

Can you really trust a page with the title, "Power Supply Decoupleing?"

Well it makes sense to me.

Let me guess,the guy beat you out of a multimillion dollar aero-space contract(;