Hi Aaron,

 A few considerations I have learned and adapted when considering noise propagation in common substrate technologies.

1.  High Frequency noise likes low inductive paths.
    This means that high frequency noise does not want to travel long distances from the sourcing power domins to sinking ground domains.  Keep the loops small by designing well coupled power and ground domains.  By this, I do not mean merely including lots of decap, but by routing paired power and ground planes/busses as close together as possible.  By placing paired power and ground pads & pins adjacent to each other when possible.  This general coupling will help keep electromagnetic fields to a minimum and high frequency noise currents will be loathe to travel too far from these coupled paths because the inductance will start to present a very large impedance.  Even in an IC substrate, electric fields can remain tighly coupled to their respective domains, if power and ground domains are very tighly coupled.  This is however a frequency dependent tactic.

2.  It is much easier to control noise at the source than it is to capture it at the destination.
    Figure out which blocks you expect will generate the most substrate noise and focus on using the P+ tap/Guardring technique on those blocks.  Determine the noisy devices and nodes and guardring those as best as possible.  Also it is not sufficient to guardring the whole block, you need to guarding the devices as close to the diffusions as possible.  For many of the noisy devices, turn the body contacts in guardrings.

3.  Practicle use of guardrings
  So many people screw up guardings, so avoid improper guarding connectivity. Example, I have seen guardrings 5 um wide with many rows of contacts all strapped in metal 1.  Then in one far corner of the whole guardring, two small vias are placed and a long 1 um wide, 100's of ums long, Metal 2 route is run to the pad connected to the guardring.  That metal 2 route alone presented >50 ohms of resistance, effectively eliminating the guardring.  If you really want the guarding to present a very low impedance, so it can easily pick up noise carriers, for heavens sake strap that ring HARD to the appropriate power domain!!!  Keep the whole path to the lowest resistance/impedance possible.


There are many other tips and techniques that can be used but I think these are good starters, at least from a very generic perspective.

Stephen
www.srftechnologies.com