What is the right cap type to use here?

[Infraviolet]

What is the right sort of capacitor to use in RC oscillators and/or LC resonance tanks so as to have a good precision on what frequency comes out of the oscillator and/or what frequency the LC is resonant at?

I've typically always used ceramic 0603 MLCCs for capacitors in PCB designs, but I note the tolerance stated is often about +/-10%, pretty wide. And while for some values more precise tolerances available, for most values when I searched on distributor websites once filters for the capacitance and the form factor were enabled, I only got options of +/-10% and +/-20% and +80%,-20% for most capacitances I searched for. Some values gave a +/-5% option, most didn't.

When I create a circuit with one such cap its behaviour seems to stay stable over time, an RC circuit doesn't vary every few cycles between different frequencies in the range or anything crazy like that. I'm guessing the wide tolerance is saying that any given cap will keep a consistent capacitance in a variety of circumstances (voltage, temperature, over months and years of time...)? But that the MLCC manufacturing process has a fair bit of variability inherent?

Is the +/-10% a rare worst case scenario for how bad they can get and still be sold as the same model, or can one reliably expect most of the caps of a certain model manufactured to differ from notional capacitance by that much?

I'd like to be able to get accuracies of perhaps +/-1% or 2% on the frequencies of oscillation/resonance in the RC and LC examples.

I'm asking this in regard to capacitors in the >100pF to <10nF range, used in cirumstances where the voltage can apply either way, so non-polar is necessary.

Is there another type of capacitor I should be getting instead for these use cases? Hopefully still available in 0603 sizing?

Thanks

[Benta]

For stability and minimum loss, PPS is the choice today.
But they are significantly larger than ceramic. The tradeoff is left to you.

[TimFox]

"MLCC" is a description of the internal construction (multi-layer ceramic capacitor), not the dielectric.
They are available in "good" dielectrics (C0G), mediocre dielectrics (X7R), and crummy dielectrics (Z5U) (ranked for use in resonant circuits), among others.
C0G capacitors are available in lower capacitance values than the others, but nowadays 0603 C0G 10 nF capacitors are available (more expensive than X7R).
C0G/NP0 has excellent temperature stability (+/- 30 ppm/K) and excellent loss (high Q).
Plastic films have lower temperature rating than ceramics, important for SMT soldering.
Polypropylene film capacitors are excellent, with very low loss, but somewhat worse temperature co-efficient (-250 ppm/K), usually TH (leaded), and harder to find in low capacitance values like 100 pF.
PPS (polyphenylene sulfide) film is available in surface mount, and has similar quality to C0G ceramic, but is only available in larger capacitance and case sizes, since otherwise one would use C0G.
Kyocera PPS capacitors start at 1000 pF and 1206 case size.
Tight-tolerance capacitors are available, but naturally they are expensive.
If you have a way to measure frequency, you can use two capacitors in parallel:  measure the frequency with only a larger one in place (for a higher frequency), then add a smaller one to reduce the frequency.
The tolerances quoted in the catalogs are the range of capacitance that comes out of the production line:  temperature drift and voltage-induced changes (for crummy dielectrics) are over and above that tolerance.

[Infraviolet]

When I've bought SMD caps before I never paid attention to anything but form factor, capacitance, making sure the voltage rating was >2x 5V (or greater than 2x 12V for the rarer times I need a cap in a circuit of that voltage) and sometimes capacitance tolerance. I think what I have to hand is a mix of C0G, X5R and X7R.

Looking on Farnell for example
https://uk.farnell.com/w/c/passive-components/capacitors/ceramic-capacitors/smd-mlcc-multilayer-ceramic-capacitors?capacitance=100pf&ceramic-capacitor-case=0603-1608-metric-&dielectric-characteristic=x7r
Almost all the X7R's for a particular capacitance and form factor are +/-10%

And all the 1% tolerance ones are C0G
https://uk.farnell.com/w/c/passive-components/capacitors/ceramic-capacitors/smd-mlcc-multilayer-ceramic-capacitors?capacitance=100pf&ceramic-capacitor-case=0603-1608-metric-&capacitance-tolerance=posneg-1pc

For less "round" values, but still in the common series's of capacitor sizing, there are less or no options with 1% tolerances.

Assuming an ambient temperature range during operation of 0 celsius to 40 celsius, and only milli-amp scale currents flowing during charging and discharging, is a C0G MLCC about the best choice one can make for oscillator timing controlling capacitors in the >=100pF to <10nF range?

[TimFox]

Short answer:  yes.
If you need an exact value, not available from stock, you can always parallel capacitors.
As a practical matter, if you need better than 1% precision in initial value, you need to use a capacitance meter (LCR or bridge) to select capacitors or add a variable capacitor ("trimmer") to the circuit.
The choice of dielectric is crucial for a critical application.
For tuning and timing, C0G is required for ceramic SMT.
For bypass and similar applications, X7R and others can work.
Note that the "crummy" dielectrics such as Z5U will change capacitance drastically with applied DC voltage.
For "crummy" dielectrics, the voltage rating indicates the manufacturer's guarantee of a safe voltage to apply without smoke;  it should not be assumed as a recommended working voltage.

[wraper]

For bypass and similar applications, X7R and others can work.
Note that the "crummy" dielectrics such as Z5U will change capacitance drastically with applied DC voltage.

X7R will drop capacitance as well under DC bias. Quite extremely for high capacitance rating for their geometrical size. Z5U on other hand is extremely dependent on the temperature as well. 2% and 1% tolerance is quite common for C0G (unless capacitance is very low), of course for higher price. For X7R and other type II ceramics tight tolerance makes no sense due to multiple factors affecting capacitance, including ageing. Actual capacitance tolerance is much worse than rated, as it's rated in out of factory condition. After a few years they will lose like ~20% on average (even without any DC bias and at room temperature), reflow will reset that and capacitance loss will start again. 

[TimFox]

Another quantitative question:
The Q value Q = X_s/ESR at a given frequency also varies considerably with dielectric.
If you have mA of current through the part, it is useful to estimate how much power will be dissipated in that ESR (making sure you know the Q at your frequency).
For C0G, roughly Q > 1000, relatively independent of frequency.
For X7R, the range of Q is roughly 20 to 40 at 1 kHz measurement frequency.
A good manufacturer's description of X7R:  https://us.rs-online.com/m/d/a24bb5190bea413b2ff164816f55e87c.pdf
(It also shows variation of capacitance with voltage.)