smallest 0603 cap

I have some 0.1 pF caps in stock too, primarily for adjusting the transfer functions of high-Z TIAs.

You could maybe make a poor man's 3-terminal cap using two of those in series--the capacitance to ground from the midpoint would reduce the end-to-end capacitance.

Cheers

Phil Hobbs

One end of my cap will be ground already.

I could just use PCB capacitance, but that's hard to tune. I'd just have to get it right first try.

Below 0.1 pF, maybe I can use a 10 meg resistor, which is around 0.04

It's already not that easy to get less than 0.1 pF to ground from a PCB pad, let alone a trace.

Cheers

Phil Hobbs

Yeah, but the usual trim techniques still work; RG174 is 30 pF/foot, so you can get your 0.1 pf by soldering a short length onto the board, and then with flush nippers, cut it off.

0.04 inches for 0.1 pf. How does that work exactly?

That's why it takes flush nippers. :-)

Yes, but when you look at the tolerances are you really gaining much?

John

That's why ADS or other CAD systems have these capacitor-less capacitors with interdigitated layout models that can be tuned with a scalpell if really needed on the hardware board. A 0201 cap that swims around during soldering?

-------- | --------|

-----*------- *-------- | --------| --------

Gerhard

I invite you to try getting the fringing capacitance of an actual piece of RG-174, connected to a circuit, to be that low. Show your work. ;)

(The capacitance per unit length only applies when the fringing capacitance is negligible.)

Cheers

Phil Hobbs

But, because this is a trimming technique, there's an end correction BEFORE the snip as well as after. At least, there is until the two ends of the cable coincide...

I'm not concerned with high frequencies much, so I might apply a negative capacitor first, swamp it with the long-RG174 trimmer, then trim down to get near zero. Negative impedance converter, you know...

Yes and no. Given invariant fringing capacitance, changing the length incrementally does give a nice predictable delta-C. Trombone line is good that way down at low frequency, for instance.

This is because, in a nice long piece of coax, the E field inside is purely radial almost everywhere, and any departure from the pure TEM mode at the ends dies off exponentially, roughly as exp(-2 pi L/r), where L is the distance from the open end and r is the radius. (That's a consequence of Laplace's equation, and is also why perforated metal makes good electrostatic shielding.)

However, John asked for a small absolute capacitance. A picofarad or so of end effect doesn't fit that bill.

Plus when the coax gets too short, the approximation you're relying on starts getting inaccurate because (a) there are varying axial fields throughout the length of the coax, and (b) the fields at the two ends interact.

Cheers

Phil Hobbs

And I want something that can be manufactured, not a hobby fiddle thing.

Something like this:

It needs extreme CMRR to work. If C1 is just a resistor parasitic, I can make C2 a bit larger and the CMRR error becomes positive feedback, which is OK in moderation. High voltage Schmitt trigger.

I could just use a digital isolator to drive the GaN fet, but they are slow. The modulated ones may add jitter, too.

Of course, detonators probably don't care about a bit of jitter.

Well, not repeatable jitter, anyway. ;)

Cheers

Phil Hobbs

Not so; this (or any) trim isn't intended to hit only one absolute value, it's for test-and-adjust finagling. If you want to hit near-zero, any adjustable capacitor plus a negative offset circuit (negative capacitance) seems to me to be a suitable solution. Having a suitable length to snip on, does eliminate the concern with an invariant element, like the end correction.

Right. Measuring jitter requires a statistically valid number of pulses.

We used to ship CAMAC systems that went down-hole near a nuke. They shipped a bunch of time stamps up a long cable. Good replacement market.

Just drilling the hole cost a million dollars.

Okay, for something like that, just dorking it to one side a bit makes sense, assuming that the worst-case RC time constant is small enough that it doesn't distort the next transition.

I'm actually working on a proof-of-concept that needs two separate TIAs to have very closely similar phase and amplitude response out to a couple of hundred kilohertz, which requires excellent phase matching. We're only making 10 or so boards, so I'm planning to put a couple of DNP caps in parallel with the main one, to get the tolerance down from ~5% to ~0.5%. (Yes, this will require a bit of measurement and selection on each board, but there aren't very many, and we have all the stuff in stock.)

Maybe it's possible to use a dpot with a shunt cap on the wiper to balance the two sides.

Cheers

Phil Hobbs

How about a trimmer cap? Beats soldering.

If it's TIA feedback, a trimpot can tweak the effective C value.

DPOTs are usually slow with the slowness code-dependent. Real trimpots are great and easy to program.

If it can be adjusted, it can be mis-adjusted, leading to hard-to-diagnose misbehavior and reasonable-looking wrong answers. (My least favorite kind.)

Sure, if you can still get them. Those nice Murata PVA2 things are long gone.

A dpot is a variable RC network, so by putting the cap at the wiper, you change the capacitive loading of the ends. There's some built in to the dpot as well, as you say, but it varies with code too.

Cheers

Phil Hobbs

Did you look at parts similar to

At least you do not start with 5 %.

Bye