Alternatives to capacitors from American Technical Ceramics

the non-idealities of the cap, inductance etc are almost all caused by the physical size. An 0402 SMT 100pF cap will be close to ideal and a 0.01uF 0402 SMT cap will actually not be that much worse because the inducatance is about the same.

Are you working from below 1MHz up to 3 GHz and that is why you need 0.01uF

Mark

I've tried the expensive DiLabs super-wideband caps and they don't seem to be different from a cheapie, at least in my TDR measurement bandwidth, about 15 GHz.

The rotated (0306 type) caps are even better in low-impedance, higher-speed situations.

These are 2.2 uF caps, 6 cents each from Digikey. An 0306 or 0508 might bridge a 50 ohm trace nicely.

Yes, my frequency range requires 0.01uf to 0.02uf. Anyway, the manufacturers' impedance plots I've seen indicate that the impedance of the standard chip cap can exceed 10 ohms at over 1 GHz - I want to avoid that.

ance up to 3 GHz. I'm using it to AC couple the RF input of a receiver. Ame rican Technical Ceramics manufactures capacitors (e.g., the 550 series) tha t have excellent performance up to 40 GHz. Of course, this comes at a cost: they are about 100 times more expensive than a standard x7r cap. I would l ike to find something that is less expensive but would have similar perform ance up to 3 GHz.

Thanks for the input. I'll do some bench testing to see how a standard 0402 x7r cap behaves in my application. I may also look into the rotated, low i nductance caps you mention, though my microstrip width may not accommodate them.

It seems good to not have the cap much narrower than the trace, so the current doesn't have to crowd down into the skinny cap. But, as noted,

The "wall in the sky" here (term borrowed from Chuck Yeager's breaking the sound barrier) is around 100 ps, which is 3.5 GHz. Things are fairly easy at 100 ps, but get harder fast.

Depends on capacitance you need. Take a ceramic cap and partly wrap copper foil around it "C-shape,cap in middle) and solder one lead to foil. The copper foil acts like a shorted turn to the internal "U"-shaped inductance of leads and capacitor.

It's not a property of the capacitor, it's a transmission line property. As long as it's going over ground plane with the correct trace width, it should be fine.

Tim

Tim, you'll need to elaborate on your statement. ESL is capacitor dependent.

ESL depends on size and shape, not on price and application domain, despite what some manufacturers would want us to believe.

If your capacitor sits over a gap in a transmission line and has the same width as that transmission line, chances are that its ESL is nearly invisible. That got said several times over, by now.

Jeroen Belleman

make your own broadband cap? well at least the high frequency part.

Well,the old ceramic caps were round, and with the leads and the way they terminated inside,made for a nice resonant circuit. Plopping on top of a trace does help like i mentioned, but the copper "C"-shaped "shield" does wonders.

Yeah, THT discs form a loop, where you have a little lead inductance (more accurately: a short, high impedance transmission line up from the pad), up to a disc that's more or less a blob of metal at RF, and back down again. So it's not just a single loop, but it's twofold or worse, because you get series inductance loss from each pin, plus parallel [stray] capacitance loss. It's a tee filter! It's also slightly dispersive, because a small part of that stray capacitance will couple into the traces (in and out), as well as (predominantly) the ground plane.

And also from whatever the lead pins themselves do: microstrip entering a pad from the top will have a short stub extending through the board (the pin entering the top and exiting the bottom), with an annular ring on the end, clearanced to the ground plane. (Top side ground plane and bottom side microstrip is probably slightly better, since it avoids most of the stub, but the layer transition kind of sucks, and you still don't fix the geometry of the capacitor, which is the majority of the problem.)

You could also bend it over so it lies flat against the board, which can help reduce inductance a little (or more accurately, keep its impedance closer to system impedance), but now stray capacitance is way higher (the disc against the ground plane or board surface acts like a big blob of low impedance transmission line).

If only you could smush that big fat disc closer to the board, and make it more rectangular, so the whole thing tends to look like a constant impedance transmission line... :)

As for shielding around a chip cap, The nice thing about that would be, assuming it's close enough to touch (ideally, perhaps: the end cap metallization itself is extended up over the body, so they nearly meet in the middle), you increase the capacitance even further, including at high frequencies: the fringe field from the closest edges will carry high frequencies quite effectively through the dielectric. And remember, we're talking k >> 10, so it's not a terrible conductor at high frequencies.

The un-nice thing is, I don't see that you'd be able to apply shielding very effectively, in a measurable way, just by poking copper foil around, or whatever. If you have good solid measurements, that would show that pretty well.

But there are other ways to crack that, too. For example, on a two layer board, microstrip (or GCPW) tends to be rather wide (50 ohms is what, 60 or 80 mil?), so a number of very small and low-profile chips could be stacked in parallel. DFM doesn't really approve of, say, three 0402s stacked right next together, but it could be done, even if by hand.

Not so easy on multilayer microstrip, because you're probably dealing with

Tim

ance up to 3 GHz. I'm using it to AC couple the RF input of a receiver. Ame rican Technical Ceramics manufactures capacitors (e.g., the 550 series) tha t have excellent performance up to 40 GHz. Of course, this comes at a cost: they are about 100 times more expensive than a standard x7r cap. I would l ike to find something that is less expensive but would have similar perform ance up to 3 GHz.

Well, I just did a bench test, comparing a $0.05 0402 0.01uF cap and a $5.0

Darol Klawetter

Am 17.12.2014 00:27, schrieb Darol Klawetter:

We had excellent results with 0402 X7R 100nF (IIRC AVX) as coupling capacitors in 10 Gbit/s XFP transceivers. 10nF were too small and produced 1 bit error/night of test time for just one polynom. Took some time to find out.

regards, Gerhard

For porcelain caps: I think the largest porcelain cap they make is 1000 pF.

Methinks that you may be doing something wrong if you need porcelain caps for the RF coupling cap in the front end of a receiver. Assuming a 50 ohm input, the difference in ESR between a porcelain and ceramic cap isn't enough to have an effect on any of the specs. I use low ESR porcelain caps in the transmitters, where the high RF current can easily cause an ordinary capacitor to overheat. What do you mean by "similar performance"?

Why a 0.01uF input cap? 0.01uF into 50 ohms is: f = 1 / (2 * Pi * R * C) f = 1 / (2 * 3.14 * 50 * 0.01*10-6) f = 320 KHz If your receiver needs to hear down to about 500 KHz, 0.01 uF is a good choice. However, if your operating range is higher in frequency, I suggest you pick a smaller value cap.

Also, watch out for the type of dielectric used. Some are microphonic, such as barium titanate used in some MLCC (multi-layer ceramic caps). These are a really bad idea for receivers.

You may have a DC component problem what they used to call baseline wander. Is there a specified limit to the DC component in your data stream? If not you May need DC coupling to do it right, Mark

• You forgot to add in the "C" shaped foil around it; that fixes the majority of the ceramic disk caps problems..

• I think that not only is not needed, but would be counter-productive. Pick a chip with width dimensions as close to microstrip trace dimensions as possible and as thin as possible to mimic the trace thickness.

Bit pattern sensitivity??

Yes, one polynom had some silly long runs of 1 or 0, several similar in a row. With the real time scopes of today it would have been caught in one test run.

regards, Gerhard