I don't like tants across power rails... they tend to explode. I do use them across linear reg outputs, because the ESR is often just right, but I derate 3:1 on voltage.
Aluminum polymers are nice, very low ESR, no dry-out, no detonation mechanism. Prices are getting more reasonable.
--
John Larkin Highland Technology, Inc
picosecond timing precision measurement
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
an smsp draws ~constant power, lower input voltage draws higher input current, with a bit of bad luck the increased current lowers the voltage increasing the current etc. etc.
It's characteristic of all type 2 dielectrics. The denser ones (Y5V, Z5U..) just do it worse, which is to say, you're really just cheaping yourself out of more energy storage.
(The higher k means lower energy density: e = D^2 / (2*e_0 * e_r) D is the electric analog of B, and E that of H. Thus, where B = mu*H for magnetics, D = epsilon*E for electrics, and where e = B^2 / (2*mu), the same is true with D and epsilon.)
The sad part is, modern X7Rs (historically a preferred choice, because of the only *moderately* high e_r) exhibit about as much voltage dependence as any other type, because they're better at making thinner, more reliable layers.
The solution is not simply to get a higher voltage rating, because the voltage rating is for breakdown. (Just as a ferrite bead's rating is thermal, not magnetic. Seeing the similarities?) The only solution is to buy physically larger parts (still with a relatively high voltage rating), so that there's more dielectric.
Or buy tantalum, which is just as expensive but catches fire, or aluminum, which expires after a while (polymers have ESR as low as ceramics, so are an excellent substitute, while electrolytics kind of just suck, but both are subject to aging).
C0G has the highest energy density (because guess what, low k!), but only becomes useful in high voltage ratings (>250V), and is impossible to get in massive values. You're kind of stuck if you need fast energy storage in a smaller package than electrolytics.
Typical behavior is that energy storage increases linearly, in the saturation region. So, it still stores more energy with increasing voltage, but not nearly as much as, say, a C0G (which remains parabolic until it physically explodes).
The behavior transitions from parabolic, at the saturation knee. Which you can find in the datasheet (well... when provided). This should be enough direction to estimate the behavior of a given cap.
1uF at 20V is a physical impossibility* in 0402. You'll incur serious inrush or startup speed problems doing that.
(*For a much less impossible, linearity-motivated definition of 'impossible'...)
Interestingly(?), electret caps are beginning to be introduced. These were featured in the Google Little Box design:
Do it all the time. The big thing is to make sure there is enough input capacitance on the supply. Of course, this isn't unique to cascaded supplies. Insufficient input capacitance can cause all sorts of EMI issues, even if it works.
The input of a switching power supply is negative impedance. Lower the voltage and the current goes up - negative resistance.
John Larkin doesn't think so. Looking at it very carefully, the top of the "p" isn't quite closed, so the hypothesis that it was intended to be a Greek mu (which tends to have a shorter tail) isn't entirely implausible.
On Wednesday, October 19, 2016 at 5:35:35 PM UTC-7, Tim Williams wrote: ...
That doesn't stop manufacturers advertising such components:
eg TDK 1uF 25V 0402
This was at the output of a power limited booster so inrush was fairly gentle and startup conditions were not an issue.
As I mentioned we did find a solution with newly available small form factor tantalums with a small COG in parallel. This also avoided piezo-electric effects that were being troublesome. This was in a VERY space constrained product.
Heisenberg warned everyone that any attempt to measure an attribute will diddle the pre-measured attribute. A 1V delta is worse that a 5V delta; maybe even more than 5 times worse. Use 5 milivolts to minimize the alteration of the actual tc.
Make the measurement in the actual application circuit, disturbing that circuit as little as possible, to get a reasonable idea of the tc in that application. Maybe that value will be close to data sheet spec and maybe not; there is bias dependency, time-from-startup dependency, and other possible traps to fiddle you up.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.