Hi,

It is really not that critical. I usually look up the load capacitance 
in the crystal data sheet and allow a few pF for wiring capacitance.

If the datasheet specifies 18pF, I would choose 15pF and it will work 
with compatible crystals.

Generally it is a good idea to follow the recommendations in the device 
data sheet and app notes regarding uC crystal oscillator considerations,

If you need really precise frequency, you'll need to consider 
calibrating the circuit with a trimmer capacitor.

However, in normal uC applications it is not really required. BAUD rates 
are usually not exact due to divider frequency errors and it makes more 
sense to choose such crystal frequencies that allow exact divider 
ratios.

Regards,
Gerhard

Hi Gerhard,
thanks for quick answer!

I chose 7.3728 MHz indeed to match to BAUD rates for serial connection. 
The system is supposed to be supplied by 3 V, which allows freqency of 
the ATmega328PB up to 10 MHz.
So I should not consider internal capacity of the pins at all? Formulas 
in the datasheet and in AN2519 request to do so...

Regards, Thomas

Thomas,

In my experience it is not worth the effort. Allowing 2-4pF for wiring 
and internal parasitic capacitance is usually adequate. As I mentioned 
before, I allow about 3pF for this and had never ever any issues. Unless 
you are required providing detailed design documentation, it is 
overkill. Besides, crystals have tolerances and unless you measure 
crystal operational parameters, you would be just spinning your wheels.

If your crystal connections to the uC pins are as short and direct as 
practical, you are done. It is usually a good idea to put the load caps 
as close as practical to the oscillator pins using adequate trace width, 
as much of the oscillator operation RF current passes through them. The 
traces to the crystal can be as thin as convenient to minimize parasitic 
wiring capacitance. Also avoid groundplanes on the other side or 
internal (multilayer) between crystal pins and load caps. But the 
reality is not as critical. The AVRs are quite cooperative when it comes 
to implementation.

Just pay attention, what the datasheet specifies and match a crystal to 
those requirements and allow for wiring parasitic capacitance.

Regards,
Gerhard

Hello Gerhard,
once again many thanks for prompt and very valuable information!
I will try 15 pF, as on other places on the PCB they are used anyhow.

Best regards,
Thomas

Hello Thomas,

You are welcome. With 15pF you will be in the sweet spot.

Keep also in mind, that the internal functioning of the oscillator 
depends on those capacitors more than the crystal. The crystal 
manufacturer uses the specified load capacitance to calibrate the 
crystal parameters.

Depending on the gain of the oscillator stage and the equivalent crystal 
series resistance you will find, actually, that the circuit will operate 
over a wide range of C values (10-50pF). But the crystal accuracy and 
adherence to the stated crystal parameters will only apply at an actual 
nominal load capacitance. This is normally hard to achieve in the 
reality.

Keep also in mind, that typical computer crystals have a manufacturing 
tolerance of 20 to 50ppm. With a series or variable trim capacitor on 
the oscillator input side, one can exactly calibrate the frequency 
within the possible range.

Be also aware, that crystals are made and calibrated to either oscillate 
in the so called "fundamental mode" or "third" or "fifth" order overtone 
mode. At frequencies under 25MHz most crystals are intended to oscillate 
in fundamental mode, although most crystals will oscillate in a suitable 
circuit also at least in third overtone at approximately three times the 
fundamental frequency. Overtone crystal oscillator circuits can be more 
critical and difficult to "tame" and require careful adherence to proper 
design practice and construction and careful testing.

(You may want to look for further information on the Internet. There is 
tons of information).

Regards,
Gerhard

Hi Gerhard,
The application is in my opinion not particularily critical regarding 
frequency. It is supposed to communicate to another device (home 
ventilation) over RS-232 serial connection at 9600 BAUD. At first I 
considered using the internal oszillator with calibration or a ceramic 
resonator, but chose the quartz crystal to be able to absorb as much as 
possible timing error on the other side by providing a reasonably 
accurate frequency on my side. With the crystal and the 15 pF load 
capacitors I hope to achieve this goal with plenty of margin.
Best regards, Thomas

Thomas wrote:
> What puzzels me, different capacity are given for XTAL1 and XTAL2, 18 pF
> and 8 pF, respectively.

You're looking at the specification of the 32 kHz crystal oscillator.

Look a couple of chapters before, for the "low power crystal 
oscillator", they specify 6 pF for both pins there.

So with an assumed additional 4 pF on both pins, you end up with 10 pF 
per pin plus your intended 15 pF capacitors, total of 25 pF per pin. As 
these 2 x 25 pF are in series, that results in 12.5 pF overall load. 
You'd like to have 18 pF … I'd rather pick 22 pF.

Overall clock accuracy is usually not such a big issue (unless you want 
to run a watch), but with the load capacitance too much off, you risk 
the oscillator not starting reliably.

Hi Jörg,
thanks for this remark.
I checked again in ATMega328PB complete data sheet and downloaded also 
most recent issue from Microchip website.
In chapter 11.3, Low-Power Crystal Oscillator, there is Table 11-3, 
providing those different 18 pF and 8 pF values for the XTAL pin 
capacity. What is strange, the top left cell of this table says "32 kHz 
Osc. Type", so indeed it seems this is valid for the low-frequency 
oscillator. So I assume meanwhile it is a typo in the document.
In which document / issue did you find the 6 pF?
Regards, Thomas

OK, Microchip first returned me the "A" revision of the datasheet. 
Indeed, revision "C" has 18 / 8 pF (and still the typo with the "32 kHz" 
in the top left corner of the table).

So with 15 pF + 4 pF stray capacitance, you end up with 37 pF in series 
with 27 pF, resulting in a total of 15 pF load for the crystal. I guess 
that's good enough given the remaining uncertainties.

So it seems everybody agrees on 15 pF for both external load capacitors, 
I will give them a try! In case it doesn't work I will come back with 
more questions...;-)
Many thanks and best regards,
Thomas

You could of course also make them asymmetric: 22 pF at XTAL2, and 12 pF 
at XTAL1.

I guess both will work.