Decoupling capacitors

n issue, so no caps was mounted. The board worked anyway due to the Stackup as Larkin explained

t time" I think it is called

Yeah, that's the guy who taught the class I took. He was very good and ver y thorough. Every significant point he made was done by explaining the the ory, showing a simulation and then measurements on a board especially desig ned for the task. That's why I have confidence in the things he taught. H e showed us the board with a cap mounted either next to the "chip", an inch from the chip or something like six inches from the chip. It made little difference how far away the cap was from the chip.... well, at six inches t here was a small dip.

The point is the guy isn't just talking through his hat. I've still got th e books around here somewhere. Very good stuff.

When you have resonances at frequencies that high you should compare the wa velength to the board dimensions. 900 MHz ~ 1 foot I believe. Was the boa rd a foot wide or maybe 6 inches? The resonances are as much between caps and the power planes as much as anything. The power planes have very low ESR and so will present rather sharp peaks if tuned with a low ESR inductan ce. The resonances between different caps are not so pronounced.

No, the two 0603 caps were directly adjacent across the same two traces. After diagnosing the problem it turned out to be the ESL of the larger cap resonating with the smaller one.

CH

There is a very simple rule about decoupling capacitors: Too many is always good Too few can lead to nightmares.

Examples :

Long time ago I noticed that when removing battery from a TV remote it was still able to send about 10 commands with no battery at all.

in parallel with it increased the life of the battery by a few years.

In a tube UHF amplifier the final stage had is 2KV anode supply decoupled by the classical mica sheet plus a 1000pF ceramic. HV supply came from a 2m long lead. When using this final stage for television I noticed some very bad echoes on the picture. It took me sometime to find that the cause was radiation at video frequencies from the 2m lead picked by the vidicon camera.

At UHF frequencies a nice trick is to use open ended quarter wavelength coaxial cable as decoupling capacitors.

Why should that matter? We're talking decoupling, not tuned circuits.

Jeroen Belleman

Why should that matter? We're talking decoupling, not tuned circuits.

Jeroen Belleman

They are different now.

Look at the Murata impedance diagrams. Rotated is better, but NFM is much better.

In 2012 I have bought some 0612, but the bag is still unopened. Their C*V product / area is worse than for 1206 DK 446-4072-1-nd

I found x2y capacitors more interesting, like DK 311-1246-1-ND originally Johanson, IIRC. They provoke a layout with more vias close together, so part of the currents in the vias cancel. Also never tested, I could not get them space-qualified.

Cheers, Gerhard

Am 11.12.20 um 00:42 schrieb Cursitor Doom:

What can be more backwards than a HP 4274A bridge? Probably more than 40 years, and you can connect the capacitors correctly, instead of using the harmless looking Hirschmann pliers.

BTW it is still quite accurate in spite of the age. I have an equally old Siemens Styrofoil capacitor, 47000 pF nominal, and it is measured as 47003 pF. On a hot summer day, would you offer.., oh sorry, it would probably just fit.

I had once an exploded electrolytic in it and replaced most of the candidates. I carefully kept away from the bridge circuits, there are zillions of things to adjust.

Cheers, Gerhard

For C*V you can have bulky ceramic some more distance away.

Yes, NFM is Muratas x2y. That is what I am taking about.

Bye

Traces??? What do you mean traces? What sort of board in 2020 doesn't have power/ground planes? Oh, was this a board designed in 1999?

So was he

a) stupid b) a genius c) just lucky

What exactly do you mean by "on-chip" in this context?

Ceramic caps in the package, not in silicon. In the course I took Lee Ritc hey was explaining ground bounce and I asked him if "in package" decoupling caps help ground bounce. He had to think about it seriously for a bit. T hen he said no. I think he is wrong. By decoupling power to ground the gr ound current surge is split between the power pins and the ground pins and so the voltage is cut, perhaps as much as in half. In addition, any pins c hanging in the opposite direction subtract from the net current surge and s o the voltage spike is further reduced.

Of course that only occurred to me years later when I was idly thinking abo ut it. lol That's the sort of stuff that you can see clearly when you are not trying.

His stuff worked, and he knew what he was doing.

I think the caps are monolithic on the chip, on the core and the bank supplies. The Xilinx BGA package is pretty thin and has no obvious bumps. I suppose a few tiny ceramics inside are possible. Our Xray doesn't have much resolution, but I might try. It's mainly a parts-on-reels counter.

Anyhow, knowing about the internal caps means I don't need many bypass caps on the board, and none need to be especially close to the ZYNQ.

I haven't seen the layout or know what part of the circuit it was for. Yes, it probably has power/ground, but its common to branch off and add extra filtering to the first RF stages, for example. But you don't do RF, so you wouldn't have much experience at that.

Most of my parallel cap woes have involved LDO output caps. Some LDOs are _really_ sensitive to the ESR, which can get hard to figure out when you've got a lot of caps of various sizes in various positions. You'll think you 've done it right, stress-tested your design, then the purchasing engineer changes cap brands and the vreg starts to oscillate.

You can see some caps as well as the crystal inside the LMK61E2:

Other shots (that I think I posted before):

LTM8049 buck regulator module

LT8650 buck regulator IC:

Kintex BGAs have the internal caps, but AFAIK the lower-end parts (including Zynq?) do not. It'd sure be nice if they did.

-- john, KE5FX

I measured a ZYNQ 7020. It definitely has some sort of caps on the core and the bank supply pins.