Hi Circuiters

I recently discovered the strong applied-voltage dependence of a SMD multilayer ceramic cap's capacitance; up to ~-80% of the nominal capacitance when biased at 50V. This explains why my decoupling with 1uF 0805's wasn't as effective as I expected, since I was introducing only 200nF when de-rated. However the manufacturers don't seem to acknowledge this behaviour in the headline figures. Does anyone know of a way to search out caps that maintain most of their capacitance at 50V?

eg. RS-online have 76 products matching: Surface mount, 50V, 1uF. I don't want to read 76 datasheets but there's no high-level filter to discriminate on dC/dV. The link below is interesting but the conclusion is that generally the bigger the case, the fewer compromises have been made and so dC/dV is probably lower. Is this really the best I can do?

https://www.maximintegrated.com/en/app-notes/index.mvp/id/5527

 

I see your quandary. I have seen that application note before. My main conclusion is to use as physically large a capacitor as is practical. But you already deduced that yourself.

The author of that app note also mentions that X7R seemed to perform better than X5R. I wonder if X8R would be better still. I suspect that all that would happen is that the X8R would be in a bigger package and more expensive.

Oh, yeah, and one other conclusion: Never use Y5V caps -- ever.


I look forward for more and better answers than mine.

 

a perfect example why one should read datasheets. another issue is that this is not exactly common info in the datasheets. i tend to use Murata caps and they specify capacitance at 50% of rated voltage. i cannot think of a parametric search engine that does this though...

 

I once had to troubleshoot a design using a 12-bit A/D that converted a multiplexed, varying input signal.
The problem was, the conversion was off by several least significant bits when it first converted a small signal right after a large signal.
After several frustrating days of trying to determine the source of the problem I finally determined that it was the ceramic capacitor used as part of the A/D input anti-aliasing filter.
The capacitor was one that used a high dielectric ceramic which, as it turned out, had a large "soakage" or dielectric absorption.
It required several milliseconds to get rid of that absorbed charge after it was charged to the high signal level, which caused an error in the low voltage conversion immediately following.
Going to a low dielectric NPO type ceramic (which has a low soakage) cured the problem.

 

Similar issues when trying to use these caps in passive filters. Filters using these have much higher 3rd order products than expected - the caps are actually changing capacitance dynamically with input signal.

 

a perfect example why one should read datasheets. another issue is that this is not exactly common info in the datasheets. i tend to use Murata caps and they specify capacitance at 50% of rated voltage. i cannot think of a parametric search engine that does this though...

Yes this is also the conclusion of the link I posted. However as you say it's not well defined even in the datasheets. Only about half the datasheets I've viewed include a graph of dC/dV. The other half just say "Due to dielectric characteristics the realized capacitance will vary with applied voltage". That's it! They don't quantify it at all

 

The wikipedia article on ceramic capacitors if full of little gems like the charts below.
https://en.wikipedia.org/wiki/Ceramic_capacitor

 

I tested one of the 1uF X7R 0805 caps I'd been using (actually 2 in series so I could apply bias and use my capacitance meter). My PS only goes up to 31V but I observed a 50% decrease in capacitance. I suppose at 50V it's probably ~70-80% which is consistent with the datasheets that specify this behavior.

 

I repeated the measurements on some murata 1206 caps. Unfortunately the results are very similar

 

Hi Circuiters

I recently discovered the strong applied-voltage dependence of a SMD multilayer ceramic cap's capacitance; up to ~-80% of the nominal capacitance when biased at 50V. This explains why my decoupling with 1uF 0805's wasn't as effective as I expected, since I was introducing only 200nF when de-rated. However the manufacturers don't seem to acknowledge this behaviour in the headline figures. Does anyone know of a way to search out caps that maintain most of their capacitance at 50V?

eg. RS-online have 76 products matching: Surface mount, 50V, 1uF. I don't want to read 76 datasheets but there's no high-level filter to discriminate on dC/dV. The link below is interesting but the conclusion is that generally the bigger the case, the fewer compromises have been made and so dC/dV is probably lower. Is this really the best I can do?

https://www.maximintegrated.com/en/app-notes/index.mvp/id/5527

View attachment 130829

AFAICR: most ceramic types are a bit quirky about voltage dependence - MLCC caps are pretty small with impressive capacitances, starting off with higher capacitance might not be such an ordeal.

If that doesn't work - make room for metallised foil types.

 

... starting off with higher capacitance might not be such an ordeal.

Yes for decoupling I'll just 10x my desired capacitance to be on the safe side but also include the desired capacitance in parallel for high freq performance

 

Are the plots in post #9 also of X7R capacitors? What were the voltage ratings?

 

Are the plots in post #9 also of X7R capacitors? What were the voltage ratings?

Yes, both X7R and both 50V. The 0805s were from China; the 1206s were from RS uk

 

I guess we can see why audio people shy away from ceramic capacitors.

 

I guess we can see why audio people shy away from ceramic capacitors.

Apparently the variable voltage dependency distorts the waveform, but they're still valid for supply decoupling.

Many TV sound boards have only MLCC caps - but "Hi-Fi" rarely pops up in the same context.