short-term stability of class-1 ceramic capacitors

[[email protected]]

It appears that the capacitance of class-1 ceramic capacitors (NP0=C0G,

N750, etc.) in general suffers changes of the order 10^-4 over short
periods of time. See the p.68 of the Vishay-Draloric General
Information
on Ceramic Single-Layer Capacitors under
<http://www.vishay.com/docs/22019/cergenin.pdf>

Apparently, the inconstancy can somehow be reduced to below 10^-6. Is
that done by special processing or treatment or by selection? Does
anybody know the physical mechanism behind the inconstancy? And what
is the typical time scale of the "short-term" fluctuations?

TIA, Martin.

 

[Joerg]

Hello Martin,

It appears that the capacitance of class-1 ceramic capacitors (NP0=C0G,

N750, etc.) in general suffers changes of the order 10^-4 over short
periods of time. See the p.68 of the Vishay-Draloric General
Information
on Ceramic Single-Layer Capacitors under
<http://www.vishay.com/docs/22019/cergenin.pdf>

Apparently, the inconstancy can somehow be reduced to below 10^-6. Is
that done by special processing or treatment or by selection? Does
anybody know the physical mechanism behind the inconstancy? And what
is the typical time scale of the "short-term" fluctuations?

Maybe by baking well above 100C. Just keep in mind that a stabilizing
process can be partly spoiled by the heat during the solder process.

Regards, Joerg

 

[Pooh Bear]

It appears that the capacitance of class-1 ceramic capacitors (NP0=C0G,

N750, etc.) in general suffers changes of the order 10^-4 over short
periods of time. See the p.68 of the Vishay-Draloric General
Information
on Ceramic Single-Layer Capacitors under
<http://www.vishay.com/docs/22019/cergenin.pdf>

Apparently, the inconstancy can somehow be reduced to below 10^-6. Is
that done by special processing or treatment or by selection? Does
anybody know the physical mechanism behind the inconstancy? And what
is the typical time scale of the "short-term" fluctuations?

If the dielectric's that 'touchy' and that's problematic for you then why
not use another one ?

You're obviously looking at small values so mica might be good for you.

Graham

 

[[email protected]]

Hello Martin,


Maybe by baking well above 100C. Just keep in mind that a stabilizing
process can be partly spoiled by the heat during the solder process.

Regards, Joerg

You seem to be thinking in terms of internal stresses in the ceramics.
Stress relief is in fact behind the aging of class-2 ceramics, the
constituents of which are ferroelectric (and hence piezoelectric) -
class-2 caps show a logarithmic capacitance reduction with time. The
oxide mixtures of class-1 dielectrics don't contain ferroelectric
components, although the different crystallites will no doubt be
stressed when the fired mixture cools during production.

I will try accelerated aging or stress relief treatments as a last
resort - if selection within one lot, replacement with caps from
different lots and/or different vendors doesn't help (10^-6 caps are ok

for my purpose; the "noise" level of some 100V 2% EGPU caps made by
Philips/BC Components with N750=U2J dielectric is obviously worse
than 10^-5).

Martin.

 

[[email protected]]

If the dielectric's that 'touchy' and that's problematic for you then why
not use another one ?

You're obviously looking at small values so mica might be good for you.

Graham

Reason is: class-1 ceramic caps can be had with virtually any tempco
from
+100 ppm/°C to minus-a-few-thousand ppm/°C. Mica is always around
+40ppm/°C, I think.

I am surprised by the magnitude of the inconstancy: 10^-4 is not
automatically insignificant, these caps are used all the time, and no
other manufacturer is even mentioning the problem!

Nor have I been able to locate information on this topic on the wider
internet.

Martin.

 

[KoKlust]

periods of time. See the p.68 of the Vishay-Draloric General
Information
on Ceramic Single-Layer Capacitors under
<http://www.vishay.com/docs/22019/cergenin.pdf>

Vishay states: "For higher requirements e.g. in commercial applications, we
can supply *tubular* capacitors, the inconsistancy of which has been reduced
to a minimum."

- This makes me guess that the inconsistancy has something to do with a
mismatch in TCE (thermal coefficient of expansion) of the SMD package with
respect to the PCB material the capacitor is mounted on.
The stress and creep typically cause sudden stepwise changes.
Using better quality PCB material (polyimide, LCP, Rogers) can also help
reduce this stress. Another thing: don't forget that the PCB material
properties determine the stray paracitic capacitance. These may be in the
same order as the 'inconsistancy'.

According to Vishay their high grade parts, OR their 'tubular' parts also
solve the problem.

 

[Jim Adney]

Vishay states: "For higher requirements e.g. in commercial applications, we
can supply *tubular* capacitors, the inconsistancy of which has been reduced
to a minimum."

I downloaded the pdf file and I have to admit that this was the very
first time I ever heard of this phenomenon. It may explain why HP used
some small tubular caps in the 10MHz clock oscillators in the
frequency counters I commonly work on.

My guess would be that the thin wall tube allows for a more uniform
finished crystal structure in the ceramic.

-

 

[[email protected]]

Vishay states: "For higher requirements e.g. in commercial applications, we
can supply *tubular* capacitors, the inconsistancy of which has been reduced
to a minimum."

A footnote: "inconsistancy" must be a typo for either "inconstancy" or
"inconsistency". The former is much to be preferred in the present
context.

- This makes me guess that the inconsistancy has something to do with a
mismatch in TCE (thermal coefficient of expansion) of the SMD package with
respect to the PCB material the capacitor is mounted on.
The stress and creep typically cause sudden stepwise changes.
Using better quality PCB material (polyimide, LCP, Rogers) can also help
reduce this stress. Another thing: don't forget that the PCB material
properties determine the stray paracitic capacitance. These may be in the
same order as the 'inconsistancy'.

I agree that mounting-related stress has to be considered for SMD parts
(I am not sure Vishay make single-layer SMD parts). However, Vishay's
introductory statement on p.68 is not restricted to particular
capacitor
styles, and so refers to ceramic single-layer caps in general, making
sense primarily for class-1 dielectrics as the ferroelectric class-2
types aren't very stable anyway (their capacitance depends on AC
voltage
and DC bias, on temperature, and on capacitor age).

The introductory statement reads: "During operating a ceramic capacitor
the capacitance value may change for short periods of time." In fact, I
don't see why this shouldn't apply to multilayer caps as well,
regardless
of their mounting (leaded or SMD).

I've thought about PCB contributions: they easily exceed the order of
the
inconstancy, but their short-term variation should not.

According to Vishay their high grade parts, OR their 'tubular' parts also
solve the problem.

Rather, as far as I could see, their (electron-tube aera style) tubular
RDQL (lacquered) and RDQT (resin coated) caps are the *only* models for
which controlled "short-term stability" (or "KzK") grades are offered
(with KzK from 4 to 6). All others models are "uncontrolled" - they
aren't
graded. See the RDQL/RDQT datasheet.

Martin.

 

[KoKlust]

Rather, as far as I could see, their (electron-tube aera style) tubular

RDQL (lacquered) and RDQT (resin coated) caps are the *only* models for
which controlled "short-term stability" (or "KzK") grades are offered
(with KzK from 4 to 6). All others models are "uncontrolled" - they
aren't
graded. See the RDQL/RDQT datasheet.

Hmm okay!
So your problem is solved?

I am very curious if the Vishay guys know the science behind this
phenomenon. Did you ask them already?

 

[[email protected]]

Hmm okay!
So your problem is solved?

I am very curious if the Vishay guys know the science behind this
phenomenon. Did you ask them already?

Indeed I contacted them using the e-mail address given at the bottom of
the single-layer-capacitor pdf files. I asked them to furnish (or point
me to) more detailed information, in particular as to the time (or
frequency) scale of the fluctuations. I got an automatic forwarding
message right away on Monday morning, but nothing since (and I'm not
holding my breath).

True, my problem would be solved if I had RDQL/RDQT caps with KzK=6. I
don't like their footprint though, and haven't looked into availability
of the different tempcos. Instead, I am thinking of checking various
class-1 caps using a simple setup (what Vishay would call a "very
sensitive measuring device") like this:

The capacitor under test is connected via a large resistor (say 22Mohm)
to a film-capacitor stabilized DC voltage of around 10V, the capacitor
voltage amplified by something like a TL071 configured for a gain of
100 or 1000, and the noise near f = 1/(2*pi*R*C) observed on a scope.
This should allow to probe dC/C = 10^-6 for 100pF caps, the limiting
factor being white noise from the 22Mohm (600nV/rtHz, smaller resistors
would mean that one sees higher frequency components only), and the
input current noise of the TL071 (about 10fA/rtHz at 1kHz) multiplied
by 22Mohm.

Perhaps interested people could share their results on the group?

Martin.

 

[KoKlust]

frequency) scale of the fluctuations. I got an automatic forwarding

message right away on Monday morning, but nothing since (and I'm not
holding my breath).

The capacitor under test is connected via a large resistor (say 22Mohm)
to a film-capacitor stabilized DC voltage of around 10V, the capacitor
voltage amplified by something like a TL071 configured for a gain of
100 or 1000, and the noise near f = 1/(2*pi*R*C) observed on a scope.
This should allow to probe dC/C = 10^-6 for 100pF caps, the limiting
factor being white noise from the 22Mohm (600nV/rtHz, smaller resistors
would mean that one sees higher frequency components only), and the
input current noise of the TL071 (about 10fA/rtHz at 1kHz) multiplied
by 22Mohm.

Wow, this is a completely new way to measure capacitance for me!
Did you ever use such a setup successfully? Up to ppm scale?
I'm afraid I cannot follow how you arrive at a ppm capacitance measurement
only with a RC, amplifier and scope.

Didn't you forget to mention something like a demodulator / 72 Hz bandpass
filter / rectifier? Else I don't see how you can make picofarads from the
noise on the scope screen. Reading the amplitude of the noise gives
peak-to-peak values that get as large as you want, if you wait long enough.
And even with a spectrum analyzer it would take a long time to get anywhere
near 1 ppm.

Well anyway you probably have an easy explanation.

Best regards,

Marco

 

[[email protected]]

Wow, this is a completely new way to measure capacitance for me!
Did you ever use such a setup successfully? Up to ppm scale?
I'm afraid I cannot follow how you arrive at a ppm capacitance measurement
only with a RC, amplifier and scope.

Didn't you forget to mention something like a demodulator / 72 Hz bandpass
filter / rectifier? Else I don't see how you can make picofarads from the
noise on the scope screen. Reading the amplitude of the noise gives
peak-to-peak values that get as large as you want, if you wait long enough.
And even with a spectrum analyzer it would take a long time to get anywhere
near 1 ppm.

Well anyway you probably have an easy explanation.

Best regards,

Marco

Well ... the idea is not to accurately measure capacitance, but to
roughly grade ceramic caps according to their short-term stability - by
converting capacitance fluctuations into voltage fluctuations: 1ppm *
10V = 10µV. Surely one can decide if there is 1mV or 100µV or 10µV
of
noise without waiting forever? But I haven't tried it yet, as I am
currently working on something else.

As some people may have problems distinguishing noise components around
72Hz (for R = 22Mohm, C = 100pF) on a scope by eye, a bandpass would
come in handy; 6dB damping per octave above and below the observation
frequency can easily be implemented in the TL071 feedback path.
Alternatively, one might directly observe the noise spectrum in the
frequency domain, using a digital scope ("spectrum analyzer").

Note that the RC combination at the OpAmp input already provides for
damping of both the resistor noise and the input current noise above
72Hz.

You never stop learning!

Martin.

 

[KoKlust]

* cut *

Didn't you forget to mention something like a demodulator / 72 Hz bandpass
filter / rectifier? Else I don't see how you can make picofarads from the
noise on the scope screen. Reading the amplitude of the noise gives
peak-to-peak values that get as large as you want, if you wait long
enough.

Well ... the idea is not to accurately measure capacitance, but to
roughly grade ceramic caps according to their short-term stability - by
converting capacitance fluctuations into voltage fluctuations: 1ppm *
10V = 10µV. Surely one can decide if there is 1mV or 100µV or 10µV
of
noise without waiting forever? But I haven't tried it yet, as I am
currently working on something else.


OK, thanks for the explanation. Maybe for a rough classification, your
method will work. But I'm afraid it runs into problems on the 1ppm stability
measurements. 1ppm is really very very hard to measure.

Your setup can be improved a lot if you leave out the 22meg resistor,
replace the 10V dc source by a triangular (or square) wave generator giving
several volts of amplitude, and demodulate the voltage at the opamp output
synchronously to the wave generator.

You will also want to use a lower noise (order <4nV/sqrt(Hz)) opamp, like
opa627.

I think that gives you a much more robust and reliable stability measurement
tool.


Well I am very interested to hear about your results as soon as you have the
setup up and running!

Best regards,

Marco

 

[[email protected]]

OK, thanks for the explanation. Maybe for a rough classification, your
method will work. But I'm afraid it runs into problems on the 1ppm stability
measurements. 1ppm is really very very hard to measure.

Your setup can be improved a lot if you leave out the 22meg resistor,
replace the 10V dc source by a triangular (or square) wave generator giving
several volts of amplitude, and demodulate the voltage at the opamp output
synchronously to the wave generator.

You will also want to use a lower noise (order <4nV/sqrt(Hz)) opamp, like
opa627.

I think that gives you a much more robust and reliable stability measurement
tool.


Well I am very interested to hear about your results as soon as you have the
setup up and running!

Best regards,

Marco

Actually, I'm proposing to do with class-1 ceramic capacitors what one
usually does with a condenser microphone.

I hope to be able to try it next weekend and will let the group know
the
result, positive or negative.

Martin.

 

[[email protected]]

KoKlust wrote:
[...]

Your setup can be improved a lot if you leave out the 22meg resistor,
replace the 10V dc source by a triangular (or square) wave generator giving
several volts of amplitude, and demodulate the voltage at the opamp output
synchronously to the wave generator.

You will also want to use a lower noise (order <4nV/sqrt(Hz)) opamp, like
opa627.

I think that gives you a much more robust and reliable stability measurement
tool.


Well I am very interested to hear about your results as soon as you have the
setup up and running!

Best regards,

Marco

Actually, I'm proposing to do with class-1 ceramic capacitors what one
usually does with a condenser microphone.

I hope to be able to try it next weekend and will let the group know
the
result, positive or negative.

Martin.

I've now done the experiment for various class-1 ceramic capacitors
(EGPU made by Philips/BC Components) around 100pF. They were charged to
5V DC via 10Mohm, and the signal was picked up by a TL071 configured
for
a gain of 1000, with 6dB per octave damping below 1/(2*pi*R*C) = 160Hz
(the feedback elements were 1Mohm between OUT and IN- and 1kohm in
series with 1µF between IN- and ground). Electrostatic shielding was
required to prevent 50Hz pickup.

No obviously capacitor-generated noise contributions were found on any
capacitor under test. The noise with ceramic capacitors was
indistinguishable from that with film capacitors and agrees with the
expected resistor noise of 400nV/rt-Hz (with 10nV/rt-Hz at 1kHz, the
current-noise contribution from the Tl071 should be somewhat below this

level).

Thus, if class-1 ceramic capacitors exhibit capacitance fluctuations of
the order 10^-4 to 10^-6, the frequency spectrum must concentrate below
160Hz. If the fluctuations proceed in sudden steps (i.e. with
sub-millisecond rise/fall times) - as one might perhaps suspect -
individual steps for the tested capacitors must be below a few times
10^-6.

In order to learn more, the experiment should be repeated with R=1Gohm
and an AD820 (about 2pA bias current, about 18fA p-p from 0.1Hz to
10Hz)
in order to probe frequencies down to 1.6Hz.

Martin.