CMOS Crystal Oscillator

Introduction

I'm currently looking at doing some digital audio stuff with an FPGA. Ideally, I need to be

working with a clock that's a multiple of one of the common audio sample rates (48kHz, 96kHz, or

192kHz). 12.288MHz would be a good choice, being a reasonably high multiple of all of those, but

it's an awkward value to generate using an FPGA's PLL: on my Lattice Brevia 2 board, with its XP2

FPGA and 50MHz oscillator, the closest I can get to 12.288MHz is out by 0.2%. The DAC part that

I'm using would work with that [the PCMxxx part that's doing the output conversion does more than

just a simple conversion and needs to be able to work out which sample rate is in use, but the

datasheet allows for 4%, so it would function ok], and if I were to base musical note frequencies

on it you wouldn't hear the difference, unless you had exceptionally good perfect pitch, but even

so I'd like to be working with 12.288MHz, so I'm going to need a crystal oscillator. I don't have

an oscillator, but I do have a couple of crystals of the right frequency, so this blog is about me

building a simple oscillator to drive the FPGA.

The Circuit

I found a really nice application note[1] from Fairchild that covers what I want to do and used

that as a guide. My design is loosly based on the circuits in the app note. I'm being quite sloppy

with all this - don't copy it for a product design. It's ok for one-off experiments with an FPGA,

though.

I don't have any 'HC04 inverters (that I can find), so I've based it on a couple of the NAND gates

in a 74HC4011.

Here's the circuit. I'm just guessing that the crystal might be cut for a load capacitance of

something like 30pF. So the load caps might have been better as 56pF, but I had some 68pF, so went

with those.

The Prototype

Here's what my prototype looked like on a plug-in breadboard

Fortunately, it started up and ran nicely.

Here it is on a 5V supply:

and here on 3.3V:

It's less symmetrical on 3.3V, though I don't suppose the FPGA would mind. The frequency is much

more accurate on 3.3V, though still a little high, so I'll go with that one since the FPGA IO is

also 3.3V. There's a little more jitter, but it looks to be less than half a ns (can't really tell

any better than that with my scope, anyway). If it were a problem, I could always give it to one

of the PLLs in the FPGA to clean up.

This is how it behaves when the power comes up.

It starts quickly enough, when the supply gets up to about 1.2V. The subsequent ramp up that you

see is simply the smoothing capacitor in the bench supply output charging.

That's enough of that. I'll add it to my Brevia 2 board prototype and see if I can generate some

waveforms with it.

References
[1] AN-340 Fairchild