We usually also put a soft ferrite in series with Vcc as close as possible to pin 40 of the MCU, with a dedicated filter cap between it and the MCU. On a 2-layer board this is sometimes the only way to control EMI.(Blast that 20-40 ground/power arrangement).

Nice to hear, that I'm not alone with this technique, Michael!
In my projects very often analog circuitry is involved, where it becomes extremely necessary to keep digital ground return currents on the common ground connection point (connection between analog ground and digital ground) as small as possible. I then use intensive power supply filtering (performed as PI filters) for groups of digital chips, like the way you proposed.
In order to minimize ground noise even further, I try to route solid ground planes on both sides of PCB, if I use double sided PCB. I do not use solid Vcc plane, but ground planes on both sides, and lot's of vias between them, as Rob Klein demanded in one of his posts.

It's interesting to note, that only the use of multilayer board does not necessarily lead to an ideal solution in order to keep digital ground return currents from degrading analog ground routing. Only the intensive use of filtering on Vcc line keeps this amount of ground return currents to a save level. This does not mean, that it's unnecessary to use multilayer board with fast digital circuitry, not at all! But it's just not enough to automatically limit noisy ground return currents to a save level, for let's say to keep a 14bit ADC 'noisefree'.

I also agree with rerranging the capacitor location next to each crystal pin; and grounding the capacitors only to pin 20 and not to the plane (might be difficult depending on your tools.) Ground-fill under the crystal might improve emissions. If you do ground-fill, it needs to be properly anchored.

In critical situations, especially if very high clock frequencies are used, the two burden capacitors can be soldered directly across pins of microcontroller. By this any copper traces are omitted 'by design'. The crystal is then connected by as short traces as possible, just like Prahlad did it.

As you recommended, there should be a ground fill under the crystal. After reading your reply I noticed, that I have overlooked, that Prahlad has connected the fixing wire of crystal to nowhere. This should be changed: The fixing wire should be connected to the solid ground plane, or at least to a local ground plane, which is connected to ground pin of microcontroller.

It must be kept in mind, that all these currents, which flow in the crystal circuit will flow back to the ground pin of microcontroller!!! And so, there must always exist an as perfect as possible connection of involved ground pins of crystal circuit (ground pins of burden capacitors) to the ground pin of microcontroller! But not enough, also all ground return currents caused by charging of stray capacitance must be fed directly to ground pin of microcontroller. And the housing of quartz IS a source of stray capacitance, which must be grounded adequately!

With modern fast microcontrollers this grounding measure of oscillator is so important, that datasheets often recommend a tested layout...


Kai