AVR Freaks

My understanding is that load capacitance filters the higher fundamentals. The crystal has more than one fundamental frequency at which it can be excited.

https://en.wikipedia.org/wiki/Fundamental_frequency

Another way to think of those capacitors is that they reduce the gain on the higher fundamentals, so only the desired oscillation is amplified.

Too much capacitance or a reduced amplifier gain could prevent the oscillation from starting. I have never considered what added fundamentals might do, maybe strange timing results.

the capacitors serve two functions:

1) Crystals are tuned to a specific frequency during manufacture. The test fixture has a certain capacitance (that appears in parallel with the crystal). That capacitance effects the actual frequency of the circuit with the crystal in it. So, to make the oscillator frequency to be within the specified error range, you need capacitors.

2) The oscillator (the active part inside the MCU) is designed to work with capacitance on the output pin and capacitance on the input pin. 

Now for the details. The NET capacitance seen by the internal MCU oscillator consists of several parts:

a) The pin capacitance. This is the capacitance of the internal chip traces, the wiring to the pin, and the physical pin. For common ICs, this ranges from 7 pf (DIP package) to 4 pf (LCC chip carrier packages).

b) The PCB trace capacitance. This depends on the width of the trace (more C if wider), the length of the trace (more C if longer), the board thickness to the next traces or ground plane under or over the trace (thinner makes more C). For 10 mil (0.25mm) traces on the order of an inch (2.5cm) long on ordinary double sided board without a ground plane under the traces, figure 2-3 pf per trace. 

c) Solder pad capacitance. If using through hole technology on both the IC and the crystal, add 1-2 pf for the pair of pads at each end of the connecting trace. For SMD, it may be half that.

d) Physical capacitors. Value should be known.

Assuming same geometry at each end of the crystal, the four capacitances at each end of the crystal add together at each end.  The resulting two capacitors appears as if connected in series across the crystal. Thus, you should have (for minimum frequency error):

Cpar = (Cpin + Cpad + Ctrace + Ccap)/2

What happens if you use a different capacitance? The oscillator frequency may differ, slightly, from the crystal's calibrated frequency. If off by a few pf then the probable shift is generally at the ppm level. Remember. initial accuracy with "correct" capacitance is often 50ppm or more. For an 8MHz crystal a 1 ppm error is 8Hz. If the capacitance error is more than, lets say -8pf or +15pf. the oscillator may not even generate a regular or steady clock.

So, the short answer is that if you are off by a few pf, nobody will know the difference in most applications.

Hope this helps

Jim