tantalum caps

We all know the limitations of tantalum. What are the alternatives?

Derate 3:1 on voltage. Or use polymer aluminums unless you need the ESR.

Or ceramics.

A few old regulators need the esr to be stable.

Thanks. I could not remember the name polymer.

----------------------

A famous engineer( Doug Self) commented there was really nothing wrong with using tantalum caps as long as you put a resistor in series and a reverse diode across each one.

..... Phil

Here's a brief review of polymer caps:

Learning About Polymer Capacitors - The Learning Circuit

Panasonic manufactures a wide variety of polymer related caps: Panasonic Polymer Capacitors: an alternative to MLCCs & Tantalum capacitors

That was written already in the NEC data sheets as of 30 years ago. Or limit inrush current.

For a space project, to my surprise I had only to derate 1:2. But then, the only allowed tantalums had already 6 times the volume of equivalent commercial ones.

... and if you don't care about the leakage current.

This here does not work:

Given is JFET amplifier with very low noise specs. CS at AF. Source current is enforced by CCS, with

FETs were IF3602. They have an updated data sheet btw. Looks like IF have admitted to the complaints of Fred Bertoli as of s.e.d. from 15 years ago, and my observations. No longer interesting.

With polymer electrolytics, the lower 1/f corner was a few 100 Hz. And with 1/f I really do mean 1/f**4,

With Nippon Chem ALU the corner was much better, maybe 20 Hz, but still steep.

With AVX wet slug tantalum, the effect was gone at a cost of 100? per 4700uF/25V. Vishay would have asked twice that. (both @ DK)

Probably with the ws tantalum the effect was also not really gone, only low enough in frequency that it did not matter any more. The ws tant still has 200 mOhm series resistance as seen on the bridge, fits data sheet.

I was puzzled by the steepness >> 1/f. It probably takes a process that collects energy over a long time and then breaks suddenly. The longer it takes, the more energy will be released, like an earth quake.

cheers, Gerhard

The stick-slip mechanism is typically more like 1/f. Random drifts (i.e. ones that change direction aimlessly during the measurement time) look like 1/f**2.

Unidirectional drift (which is consistent over the measurement time) goes like 1/f**4 in power. You can see that from the formula for Fourier transforms of integrals, applied to the delta function successively:

delta(t) has transform 1 H(t) (the unit step) has transform -1/(i2 pi f) unit ramp has transform -1/(2 pi f)**2.

These are amplitudes, so the power spectrum goes like the mod squared, so drift goes like 1/f**4.

An electrolytic cap has a certain amount in common with a dead battery, whose cell potential is temperatured-dependent, so a slow temperature drift will quite plausibly cause 1/f**4 noise.

Cheers

Phil Hobbs

I hardly use tantalums for anything nowadays.

One nice feature of alpos is that they come in a wide range of ESR for the same capacitance and voltage rating.

To make bringup easier I often put zero-ohm jumpers in series with regulator outputs (on the far side of the feedback network). That has the useful side effect of adding a few tens of milliohms, which is often just right for stability. If not, a nice pulse-rated current sense resistor fixes it right up.

A question for the assembled multitude: Since there's nothing to dry out in an alpo, does it really make sense to rate them for time@temperature as if they were wet electros?

I mean, what's the mechanism for a 1 khr @ 105C rating being different from 10 khr @ 105C?

Cheers

Phil Hobbs

Am 30.08.20 um 15:57 schrieb Phil Hobbs:

Sorry, it is only 30 dB/decade.

The other traces mostly follow the Agilent 89441A.

cheers, Gerhard

If the temperature fluctuations have a flicker spectrum (1/f), the same mechanism will give 1/f**3.

Cheers

Phil Hobbs

Polymers are fabulous. ESR is very low and most can stand substantial over-voltage and reverse voltage; there is some variation between manufacturers.

In theory, they should last a lot longer than wet aluminum caps, although those seem pretty good lately too.

I use a 56u 25v UCC part at +20 to -6 volts and it's fine so far. I tested a lot of them at -10 for a long time.

Here's some rough data:

I used those as a bulk bypasses on a bipolar programmable power rail, into a pulse generator, so I just did some relevent testing.

The big military wet-slugs, the CS13 types, don't have the MNO2 detonation mechanism either. That price is shocking, but 10m is a big tantalum cap.

I've heard that moisture intrusion hurts polymer caps, as moisture loss hurts wet aluminums. Both need good seals.

I've posted my trick for getting LM317/1117 type regs to be happy with ceramic loads.

My recollection from a couple decades ago was that wet slug tantalums were the only space qualified electrolytic capacitors. Dry slug tantalum were considered to be the least reliable and wet slug the most reliable of all electrolytic caps - is that still true?

Hmm, maybe so. The chemical potential driving the diffusion is very different in the two cases though--AFAICT there'd be no particular reason for water to diffuse preferentially into the cap unless it were underwater.

Yeah, it's strange-looking--one would initially expect that to make it worse. Looking at the FB pin as the input of an op amp, it does provide some phase lead as well as gain peaking. With 10 nF to ground and 250 ohms from the output, it'll provide a phase lead from some lowish frequency out to about 70 kHz. A big alpo could easily drag the loop bandwidth of an LM317 down into that range.

Cheers

Phil Hobbs

It doesn't make sense, but it works fine. The cap has to be the right size. I discovered it experimentally, but it works with a couple of different 317 Spice models, too.

Just a big cap from ADJ to ground doesn't help.

An alpo or a tantalum stabilizes a 317, as long as you don't put too many ceramic bypasses on the board too.

Right, because it runs out of phase lead before the loop gain crosses 0 dB.

I often parallel aluminum electros and ceramics of comparable values at the output of switchers. The one works as a lead-lag network for the other.

Cheers

Phil Hobbs