Holy c(r)apacitor dielectric, Batman!

I needed film caps for energy storage and ceramics for the fast stuff. Both.

I gotta do some math on that but it'll

Too busy? That's a lot better than the reverse.

There is some middle ground, though perhaps that's just dreaming.

Yes, I had a dyslectic senior moment with the GMR => GRM. And I dropped a keystroke with ME4L.

Good that it works for you even with only 10% of specified capacitance.

I wonder how such a capacitor might affect the waveform of a sinusoidal AC signal when used as an RC low pass filter? I would expect it to distort with even harmonic components and result in a waveform with wide zero crossings and a pronounced peak at the highest applied voltage, although with significant phase shift.

Paul

film

ely

first,

m caps

ene

eding

That's what I have bodged in. It doesn't seem to actually need it if my calculations are right, but this adds a welcome safety margin. And we can always Do Not Populate if it comes to that (assuming this all fits on the board!).

ing

Oh, not for me. I'm perfectly happy reading, learning, and working on my own inventions (of which there are many).

I always have fun pretty much no matter what I'm working on though, and I'm always working on something. Creating things is fun.

Cheers, James Arthur

------- ?All of humanity's problems stem from man's inability to sit quietl y in a room alone.? --Blaise Pascal

Replace the numerator with the physical volume of the device and you have your crapacity.

This ain't rocket science.....actually it is, just do the basic balance of energy density. Nothing unexpected about that.

Tim

Well, quite simply the cutoff frequency varies during the cycle!

What that does to the overall response, depends on what you're looking at. Steady state AC is no longer true (it's not LTI), and some of the fundamental energy is being up-converted to harmonics.

The LC case is more interesting. When sweeping frequency up or down, the spectrum exhibits amplitude dependent hysteresis.

Pair it with a ferrite bead for truly horrific conditions. With linear L and C, power-on inrush can never reach a peak voltage more than double the supply voltage. But with saturable L and C...

Tim

That gives you a crude metric for volumetric efficiency. But who cares about VE? Most of us want to know the capacitance; how they get there is up to the manufacturer.

Energy density for each of oo ceramics seems a pointless calculation; what matters is the capacitance *actually available* in a given footprint at a given voltage. Which should be spec'd up front and center--we oughtn't have to calculate or dig for it.

Cheers, James Arthur

Here is a simulation of the R-C integrator with John Larkin's non-linear capacitor model. The model is flawed for negative values of voltage, but the positive portion exhibits the peaking behavior:

Paul

Well if you want to hold one hand over your eyes and play pin-the-solder-on-the-cap, that's up to you... ;-)

The facts are:

1. There's only so much energy density they can possibly get, for any possible formulation and degree of saturation;
2. We're probably near the theoretical limit now (or within like 50%, not the perhaps 10-20% of early chips*, let alone disc-kind**),
3. C(V) curve can be, very accurately, factored out as an independent parameter. This leaves the free parameters being volume (in other words, energy density), and the C/V tradeoff (thickness, number of plates).

That is to say, the geometry of the curve itself is independent, and you can scale that curve, X and Y for volts and farads, to fit any standard part, with a pretty small error band.

**Ditto, but some of these were /so huge/ that the effect is negligible, like a 20mm dia. 0.1uF 250V Y5P, that might be a couple mm thick, that only begins to roll off at its rating. (Probably a double consequence of poor purity and processing: low breakdown field and only modest k.)

Here's a crude comparison of two Samsung parts:

You can see the 2.2 beats the 1 (and, actually, is mostly double the value anyway, which is great, but rather shuts me up). But it's way down in the saturated region, too.

Well, it's more like 1.1 vs. 0.7uF at 5V, so I'm sure my estimate is poorly drawn. Wouldn't be surprised if the red part crosses under the green curve around 20V.

In any case, you can get more, absolutely -- but because it's down in the saturated region, C(V) is strong, and you are subject to diminishing returns. A 2.2uF cap won't be 2.2 times a 1uF cap, maybe only 1.5 times. And a 10uF cap will be even worse, maybe 2.5 times instead of 10 times. It's still more, but it's not as much as you might've thought.

Using "excessively large" caps might still work out for bypass purposes -- but the extreme change in C may be unacceptable for dynamic reasons. So, this design choice depends even more critically on the circuit it's in.

For example, the voltage overshoot, under hot-plugging inrush conditions, will be many times the source voltage!

Another example: a tight control loop, say for a high-agility power supply (which might be a class-D audio amplifier, or an AM modulator, or an electromechanical controller, or..), can't afford to waste loop stability guard-banding s***ty caps.

Tim

But that's speculating, which isn't a reliable way to spec parts.

That method doesn't work. As I documented upthread, the ceramic formulae vary widely even within the X7R offerings of a given manufacturer.

Kemet's webtool says C0603C104K4RAC (100nF 0603, 16V, 1.6 x 0.8. x 0.8mm) is -3% @ 10VDC C0805C104K3RAC (100nF 0805, 25V, 2.0 x 1.25 x .78mm) is -7% @ 10VDC

The 0603 with *lower* volume and *lower* voltage rating has *lower* capacitance loss under d.c. bias.

(The Kemet webtool itself produced those part numbers, but reports "The part number you entered is not registered in the available K-SIM part numbers" if you try entering those same part numbers directly. You have to check off the size, dielectric, voltage, and 10% tolerance selections, then it'll recommend them, plot the specs, give you a datasheet, and swear it's never heard of them. Lame.)

You've also overlooked internal geometry as an independent variable, particularly plate separator thickness.

I suspect it's the latter that has changed so much since I first plotted this phenomenon a couple decades ago. Back then X7R was nearly immune to capacitance loss under bias (at least in my measurements); now that's decidedly no longer true. The vendors' websites indicate they're using finer ceramic powders for closer short-free spacing, resulting in higher electric field strengths.

Gee, that was a lot of words, Tim!

It's not that complicated. Parts should live up to their advertised ratings or have a nice data sheet documenting variances. They should not require measuring with a micrometer, computing volume, then guessing. It would be nice if we didn't have to ask a webtool for permission on each part, but that's better than nothing.

The guys have pointed to some online tools. But it's still a chore qualifying a hundred parts to have to try each candidate in some on-line tool to find out what's inside it.

Not super helpful. But better than nothing.

Cheers, James Arthur

Spec-ulating? ...Nevermind.

Well, as I said in (1), it can always be /worse/... they might be using an older formulation or a non-maximal number of plates in either of those.

That's why I went for the large values as an example. They're much closer to maximal.

Anyway, a few percent isn't saturated. They don't even rate inductors by that amount of saturation; for ferrite-cored ones they use 10 or 20%, and for powdered iron types, more like 30%. The same should apply here, i.e., it's not "saturated" until it's over 30% reduction or so.

Which for both of those I'm guessing is in the 20 or 30V range..?

Well, regardless. "Speculating" is still an accurate description. General capacitors can be worse in two ways, which obviates my claim. They can be worse out of laziness, or by process. In the old days, it would've been the latter: the dielectric simply wasn't very good, so they used a lot of it. Nowadays, they could make a world's-shittiest-0.01uF by using one layer of very closely spaced plates in the middle of a fat-assed 1206, and it'll drop off at 20V just like the above examples.

But that's the motivation for maximal values, again; you can't have a lazy capacitor, not when it costs a buck each and has to deliver 10uF in an 0603. It still feels justifiable that one of those worsening-ways can be avoided by considering only maximum values.

Jim, an IPC standard 0603 is 60 by 30 thousandths, give or take 4 thou. You /don't/ need to mic a damn thing. Thickness varies, but it's never more than the width. That's the only free parameter to keep track of. I don't think you can even get 0805s thinner than 0603s, for example, so there's no going backwards between size codes: the volume always goes up with size!

As for data, YES! That's why I picked two Samsung parts, because they usually have plots in the characteristics sheet, accessible straight from Digi-Key, like they should be. But that, too, was rather contrived on account of the links direct to Samsung's database didn't work (404 links to your own website, thanks guys -- I noticed Panasonic parts having a lot of these too, FFS), and several of the largest values didn't have a characteristic sheet at all [yet?].

On a related subject, I suggest buying Laird ferrite beads, because they actually plot DC bias. For all their chip beads.

Tim

What makes you trust the physical specs?

It varies perhaps 2:1 in the caps I'm /still/ qualifying this morning.

But I don't want to track it at all--I'd rather not be in the business of estimating capacitor volumes. That's a questionable use of my time for an unreliable predictor of capacitance loss under bias.

Cheers, James Arthur

That's actually the germ of a good idea--capacitor-makers could specify "saturation voltage" in the same way inductor-makers spec i.sat...

Vsat could be defined as "the d.c. bias that reduces capacity by 30%."

That would be very helpful.

Cheers, James Arthur