The term "island" means "seperate from". There are certainly some cases where this is necessary. As a general rule, no one rule, IMO, can be stated for all cases.
Isolating I/O ground from processor ground may not mean the complete disconnection of the two. IOW, it doesn't have to be done with "islands". It could be done by the moat and bridge method. A connection between the two grounds is still maintained but in a restricted sense. Using multiple ground planes will cause impedance bumps in a controlled impedance board that when traces are checked with a TDR can be easily seen. This may result in unwanted reflections. My own rule of thumb is to maintain one ground plane as well as I can and as large as I can (obvious considerations taken into play) to maintain conductive/impedance integrity. I always seem to be reminding the people with which I've worked that cutting up a perfectly good ground plane should be avoided unless with good reason. That you have a good ground doesn't also mean good EMI performance either. Layout of circuits in regard to partitioning is also important. Ground stitching must be done with due consideration as you have stated. I would have be guarded about stating something in very general terms. Regards, Doug ---------- > From: UMBDENSTOCK, DON <[email protected]> > > While reviewing the EMI mitigation techniques of our engineers, a few of > us have been debating the use of ground islands on a multilayer board. > > The designs have one layer for Vcc, one for ground and 2 signal layers. > The Vcc layer is isolated into various islands. The debate involves the > segregation of grounds. > > The benefit of isolating digital from analog ground is understood. The > question is whether I/O ground should be isolated from processor ground. > Some schools of thought say keep the ground plane whole for lowest > impedance, stitching it to the chassis ground through conductive > stand-offs periodically. Some prescribe segregating I/O from processor > ground, being careful where to place chassis connection points to > prevent ground loops and ground reference corruption. > > I would be interested in hearing from the forum what your pet approach > is and why you believe it works for your situation. > > Thanks in advance for your comments. > > Don Umbdenstock > Sensormatic Electronics Corporation
