BLE frequency accuracy

[BOE]

  We found a problem with a client's legacy design.  They use two 12 pF load capacitors for the 32 MHz crystal, which has a 4 pF capacitance.  This is a somewhat unusual Abracon crystal, but within the Nordic nRF52840 datasheet specs.  We found that 2 pF load capacitors are optimal.  We are trying to minimize power, so we do like the crystal.  The 12 pF caps, as far as we can tell, were simply a mistake made by the original designer. 

  Measured oscillator frequency with 2 pF load caps is 32.00000 MHz (+/- 6 ppm) over a broad temperature range.  With 12 pF caps this averages 31.99744 MHz over five units tested.  All have low frequency.  Several prototypes have been released with the 12 pF caps.  They work fairly well.  There have been anecdotal field failures.  A few units we tested will not connect BLE at low temperature, when frequency is most out of spec.

  Some voices at the client are saying: 'but we always used that value [12 pF]; it works, so don't fix it.'   Our concern is that operating at such an inaccurate frequency puts us out of compliance with BLE, US FCC, and CE.

[Hieu]

Hello BOE,

We recommend this formula to calculate the load capacitance:

Ccap = 2*CL – C_pcb – C_pin

Where CL is the capacitance of the crystal, C_pcb is the parasitic capacitance from the PCB, and C_pin is from the clock pin of the nRF SoC.

For the HF crystal on the nRF52840, we usually roughly estimate (– C_pcb – C_pin) to be -4pF. This is assuming negligible parasitic capacitance.

The fact that you measured 2pF to work best suggests that something is unusual with the situation. It could be that the crystal traces produce too much parasitic capacitance but could also be something else.

Could you please share the details of how you measure the oscillator circuit frequency and the part number of the crystal?

It might be helpful to look at the PCB design too. If you want to share this, I can make this case private.

Hieu