I'm choosing a crystal for MK20DX256VLH7, but I can't seem to find enough information in MK20's datasheet for the oscillator.

So, what I'm interested in knowing is: What is the impact of crystals' frequency stability, ESR, and load capacitance on the performance of the MCU? Would it be acceptable to use 20 ppm, 150 Ohm, and 8 pF values?

The other question on my mind is: By what margin I can deviate from these specs without affecting the performance of the MCU? Can I choose 18 or 15 ppm? Can I reduce the ESR down to 40 Ohms? Can I select load capacitance of 4 pF?

  • \$\begingroup\$ Do you mean "crystal" (rather than "crystal oscillator")? \$\endgroup\$
    – Andy aka
    Feb 9, 2018 at 16:01
  • \$\begingroup\$ Yes, I meant a crystal e.g. FW1600010. \$\endgroup\$ Feb 11, 2018 at 7:55

4 Answers 4


The crystal in combination with the circuit on the microcontroller forms a crystal oscillator. The function of this circuit is to provide a clock for the microcontroller.

You could also use Spehro's Suggestion and use an external crystal oscillator. That combines a crystal and a circuit containing everything that's needed to make that clock signal.

It might be slightly cheaper to use a crystal instead of the crystal oscillator. However, you should follow the recommendations of the microcontroller's datasheet regarding that crystal, there mainly frequency is important. You should also follow the recommendations of the crystal manufacturer's datasheet, there the load capacitance is important.

It is not difficult to get this "right" but get it wrong and it just won't work and that will be a pain.

Also a parameter like the 20ppm accuracy is often irrelevant as crystals are by themselves already very accurate. Also the microcontroller itself doesn't care about accuracy, it would still work even if the clock is extremely inaccurate and varying over temperature and whatnot.


It would probably be better for you to choose a crystal oscillator rather than a crystal. You just buy it and it works, and is guaranteed to start and conform to stated specifications.

If you are using a crystal then you need to consider drive power, load capacitance, accuracy as well as other things. In the datasheet they recommend you to refer to the crystal manufacturer for information. The crystal manufacturer will likely be similarly unhelpful and point you back to the IC maker.

You should get the load capacitance correct. Refer to the formulas for calculating it, which I won't repeat here, it's not simply the stated load capacitance stated for the crystal, it's double that minus whatever capacitance is built into the chip or exists parasitically.

If you exceed the maximum drive power you can cause early aging or even failure of the crystal. This is more of a potential issue with smaller SMT crystals which have maximum drive power of 100uW or less compared to mW for HC-49 style crystals. Lower ESR is better. Some crystal manufacturers recommend you measure typical drive power, which requires special FET probes. You can determine an upper limit to the drive power from the ESR, but that may be too constraining. Low frequency tuning fork crystals (like the typically 32.768kHz crystal which your chip can use for one of the oscillators) usually require a series resistor to limit the drive power.

As to whether frequency drift and initial accuracy is a potential problem for you, only you know to what use you are putting the chip, so this is entirely for you to evaluate from a system point of view.

  • \$\begingroup\$ Well, the vast majority of microcontrollers use crystals (or resonators), not complete external oscillators. That is, the oscillator is partially internal and when you hook up a crystal and some passives to 2 designated pins, you get a crystal oscillator circuit. It is really just a matter of RTFM and see what the manufacturer tells you to do, then do it. And in most cases the values of the passives really don't matter that much. \$\endgroup\$
    – Lundin
    Feb 9, 2018 at 15:32
  • \$\begingroup\$ @Lundin which doesn't mean (in the most cases) that you can't simply inject a square wave on the Xin pin and leave the Xout disconnected to make it work with an external oscillator. \$\endgroup\$
    – Arsenal
    Feb 9, 2018 at 15:40
  • \$\begingroup\$ @Lundin My point is that RTFM does not lead to useful guarantees. If it fails to start (say) at the lower end of the temperature range because of reduced gm of the Pierce oscillator amplifier then you have a problem, not the suppliers because they don't have that spec on their datasheet and do not ( will not) guarantee it to work. Indeed, usually it will work acceptably for non-critical applications and maybe that's fine. OP has a relatively expensive Cortex M4 processor and an oscillator is a reasonable solution. \$\endgroup\$ Feb 9, 2018 at 15:47

If you are concerned about time and frequency sensitivity and you don't know enough about XTAL OSC design, then you definitely want a Crystal Clock chip.

They are cheap, relable and have a variety of choices on stability and tolerance.

I would choose a $1~2 (1k) TCXO chip Choose from desired ppm over temperature. 10 ppm, 5, 3,2.5,2,1 ppm.

example enter image description here

There may be volumes of books on how to make a 1 ppm TCXO by now, but circa mid 90's we made one for about $1 using special jigs to instant sort xtals and varicaps for tempco and C(V1/V2) using 25 cent crystals and a 3rd order algorithm I created for HC11 controlled 928 MHz Tx synth.


Many oscillators for MCUs are designed in a similar way. So you might find more information on another data sheet preferably from the same manufacturer. You can also look for application rapports concerning CPU clocks. I found that even if capacitors are omitted the MCU works fine. Perhaps because there are enough stray capacitances around – these usually are in pF range. But my circuits are for teaching and I can simplify a lot of things with no penalty to save time and components. But If I design for a market then I try to follow advice just to be sure.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.