# Why is a broken ground plane not as effective as an unbroken one?

I did a two layer board a few weeks ago which had a dedicated ground plane. I routed 90% of the signals on the top layer and for the last 10% I had to route them through the bottom (ground) plane.

I was told that, generally, it's a bad practice to have a broken ground plane as it is not as effective as a solid one. Why is this so?

Does this also apply to power planes? Should I only route signals through my Vcc plane as a last resort? What do I sacrifice if I do so?

• This TI paper explains pretty well ti.com/lit/an/szza009/szza009.pdf It is a relatively short read and very worth while. If you don't understand all that it is talking about feel free to come back and ask. Commented Dec 1, 2011 at 14:13

Think of the high frequency currents that are running accross the ground plane.

At low frequencies, the current follows the path of least resistance (literally). A island in the ground plane isn't much of a issue in terms of resistance. There is still plenty of copper on either side of the island so that the current can flow around it with little voltage drop.

However, things look different at high frequencies. The high frequency return currents in the ground plane tend to follow the same path as the forward currents on the other layers. This is a useful property since it minimizes the total current loop area, and thereby it radiates less and the loop is also less susceptible to incoming radiation. Islands in the ground plane force currents to go around them, which may significantly increase the loop area of high frequency currents. Looking at this another way, you can think of the conductors on the top layer as forming a transmission line with the ground plane. Island break this transmission line, which increases the impedance, which increases the voltage drop accross the ground plane.

Another effect is something known as a "slot antenna". This is the inverse of a dipole, but behaves just like a dipole for radiating and receiving. If you have high frequency current running down the length of a conductive sheet and then cut a slot in that sheet perpendicular to the current flow, you have a slot antenna. This is one reason that air flow holes in metal chassis are usually a bunch of holes, not slots or single large openings.

On a two layer board, you usually have to route some of the signals onto the bottom layer. But, you want to leave the bottom layer a ground plane to the extent possible. From the analysis above, you can see that more small islands is better than few large ones. The metric you want to strive for is to minimize the maximum dimension of any island.

I use Eagle and its auto router often for such things. In the first few routing passes I set the costs just to find a routing solution. In later passes I assume a solution has been found and now it needs to be optimized for least damage to the ground plane. To get that, I set the ground plane layer cost high and the via cost lower. That results in more short "jumpers" in the ground plane layer instead of long traces. Unfortunately Eagle still tends to clump these jumpers together, even with the hugging parameter set to 0. After the final auto route, I manually clean up the ground plane a bit. This is usually not changing the topology, but mostly separating individual jumpers from each other so that there is copper flowing between them.

Here is the bottom layer drawing of such a board:

This shows the bottom layer of our USBProg PIC Programmer. A circuit of that complexity can't be routed on a single layer, but note how there are lots of individual small islands instead of long traces or large clumps of jumpers in the bottom layer. For the most part, the high frequency return currents can still flow without too much deviation from their ideal paths.

• Could you set the clearance rules between any two non-ground objects in the groundplane to be much higher than that of ground objects in that plane or non-ground objects in other planes? Commented Jan 13, 2014 at 23:59
• @supe: I use Eagle, and it allows for clearances between different net classes. However, that is not what you want to do. Setting a high clearance between different nets will make routing on the top side much worse. I don't know of any way to have Eagle automatically de-clump jumpers in the ground plane. A negative hugging parameter might work, but hugging isn't allowed to be set negative. Commented Jan 14, 2014 at 14:07

As an analogy, imagine a thin wire a km long vs that same amount of copper folded up into a square plane. The wire has far more resistance and inductance; the ends are far enough away from each other that they might ("temporarily") be at very different voltage potentials, and it won't work as well for compensating return energy.

Breaking up the plane forces return current to detour around the breaks. That forces a longer path, and since you haven't added copper, it also makes each section of the path narrower. Since lengthy detours are the problem, the blip of a small hole in a large plane matters much less than the long gap of a slot, and even a slot doesn't matter much if no signal happens to be crossing that particular gap. Thus the old recommendation that you separate analog and digital grounds (with no signals crossing), and the modern recommendation that you keep a solid plane, but put the noise sources in different zones, so that at high frequencies, they will act as though they were separated.

And, to be honest, the push that made me enter a new answer, even though @olin-lathrop had already answered well, was to highlight the TI paper recommended by @Kellenjb in a comment. Even a quarter century on, I still found PCB Design Guidelines For Reduced EMI to be very helpful, and

section 2.2.3, Gridding to Create Planes

was what I was really looking for when I searched for questions and answers like this one. Do note that gridded planes (as used there) are not the same as hatched planes. Hatched planes are basically actual planes with many small holes, so they need the same real estate; a gridded plane is a plane synthesized across layers, and therefore tolerates breaks on any one layer.