The issue of splitting planes has been brought up on the forum before, so I thought this might be of interest to some.

I have no doubt that there are some on the forum who have circuits they've designed that were so critical they actually needed to split their analog and digital power/GND planes. And I am not going to tell you that it wasn't necessary in your case. But I will tell you that, in my experience, I have not yet find a case wherein I needed to split a plane (except to accomodate a different voltage level). Moreover, if the only reason you are splitting a power/GND plane is because the ADC/DAC manufacturer's data sheet said to, then there is good reason to try not doing so. (If you're squeamish about it, remember how many data sheets insist that all you need for a power on reset pulse is a capacitor.) And if your's is one of those applications that's just too critical and you must split a plane, then here is something you absolutely must understand when you do.

The plane of a circuit board is the path wherein current returns to its source. And no matter how circuitous a path it must take, all current will return to its source. But after it follows the trace out to a load, and passes through to the plane, what path does it follow in the plane to get back? The shortest path, a straight line? No. As with all current it follows the path of least resistance. That means that it follows as closely as possible right under the trace it went out in. Here's why. (This is also why if two ground planes are properly split on a board, they won't cross talk even though they are both connected to chassis ground).

The impedence of the trace is directly proportional to the vertical distance between the trace and the plane. Likewise, the farther the distance from any point in the the plane back to the trace, the greater the impedence will be in the plane at that point relative to the trace. Thus, the current will always try to stay directly under the trace it is associated with.

So, if you're worried about noise getting from your digital circuit into your analog circuit (or vice versa) then it is usually sufficient to simply not route the traces near each other. Splitting the plane(s) is usually overkill. But what if you must split the planes? What harm can it do? I'm glad you asked.

Suppose your trace is going happily along and it inadvertently passes over one of those plane splits. As long as it is over the plane it has a nice low impedence and everything is just peachy. Then it hits the edge of the plane. Suddenly, the distance between the trace and the plane is a lot higher and increasing. This means the impedence just shot up abruptly. Suddenly those charged particles are being accelerated (slowed down), which means they are now emmitting electromagnetic radiation (EMI, RFI, whatever you call it). In fact, if you've ever studied microwave transmission you probably recognise the description of a slot antenna (this is really how they make them, by designing a slot into the ground plane and running the trace over it). And in a case such as this the current will find a way back to its source, even if it has to go out through chassis ground and back through the other ground plane to get there. Not only can you fail EMC (ElectroMagnetic Compliance) testing but you can find yourself with a load of shared noise despite your split plane.

So, if it isn't necessary you're better off not splitting any planes. But if you must split planes, sometimes it is necessary, then you really must be aware not to run traces over the split.

The preceeding two cents have been brought to you by,

Joe