When I look at this circuit in the datasheet, it is telling me that each crystal pin is connected to ground through 10pF capacitors. This suggests to me that C2 and C1 should each have 10pF added to its values to determine the correct calculation of the crystal because the capacitors are in parallel. When I contacted microchip regarded the crystal calculation, they suggest that the 10pF is added to the stray capacitance instead. I have shown the math in the attached picture. If my calculations are correct, and their guidance of using no more than 15pF for capacitors is a joke, then I could get the crystal to work with 22pF capacitors. But if their math is right, I can probably succeed with 10pF capacitors. I tried 15pF and had no success. I assumed my stray capacitance is 3pF. The crystal pads on my PCB is 2mm wide and maybe 1cm long with 0.3mm spacing in-between. All capacitors I use are np0 type with 5% tolerance.

So who would be right? Microchip with me using 10pF caps, or me with using 22pF caps?
My crystal has a CL value of 18pF.

 

If I recall correctly it is a series resonant circuit, a Pierce Oscillator. What I do is, I match the specified load capacitance on the crystal datasheet. The input, output & stray capacitances, is the effective total load on the crystal.
Make C1 approximately = C2 so CL is close to the number on the crystal datasheet.The crystal is a very high "Q" resonator, slight mismatches in the capacitance won't have much effect on stability at startup.

 

how much qualifies as slight? like 10% of the CL or less or?

 

Plus or minus 5 picofarads I would suggest you study Pierce oscillators

 

The way you question the data sheet and the manufacturer is inspirational

 

The main impact will be the exact frequency of oscillation.

Crystals can be "pulled" slightly from their nameplate frequency by varying the capacitance.

I suffered this while trying to make clocks using 32.768 kHz crystals, use the wrong value and the oscillator freq is off.