Copper pour on RF PCBs

Click here to go to our main PCB page

Not that long ago, there was a clear division of labor between microwave engineers and printed circuit designers. These days, the lines are not only blurred, they have all but disappeared.  If you are a microwave engineer and have mastered RF PCB design, you are one of your company's most valuable assets!

Copper pour is when a copper feature is "poured" between active traces.  If your traces carry RF or high-speed digital signals, merely pouring copper between traces will actually increase cross-talk. If you stitch the copper pour to the ground plane below, you can get some additional isolation between traces, if you do it right.  What is the optimum spacing between stitch vias to minimize cross-talk?  An RF engineer, will probably tell you to throw in thousands of vias at the minimum spacing, but that is not realistic from a manufacturing point of view.  A non-RF PCB designer might use a rule of thumb that 1/8 wavelength at the maximum frequency is sufficient.  And they might tell you "I spaced the traces at 5X the substrate height, so I don't even need a ground pour".  Or they might ask, "what the heck is coplanar waveguide?" Another point of confusion is that PCB designers talk about cross talk in percent, rather than in decibels. Looking at data on a linear scale (percent) instead of log scale (dB) tends to underestimate problems... If you plot the Dow Jones industrial average from 1920 to 2020 on a linear Y axis, you might never know there was a Great Depression. Go to this website site and plot Dow data on both log and linear axes and you will see what we mean.

Your engineering job is to first step back and ask yourself, "what happens when certain traces couple to each other?" before you even worry about copper pour and stitching.  In a high-speed digital circuit, 20 dB isolation between traces will prevent a one from becoming a zero, so you don't have to put a lot of thought into signal isolation.  In a microwave transceiver, you don't want the transmit signal to get anywhere near the received signal... maybe you want 80 dB isolation. Because you'll never get 80 dB isolation even with infinity vias, you should be thinking about a lid to isolate the traces (along with a ground pour and a triple row of microvias surrounding both traces).

For May 2023 we have added a link to one of Altium's many videos on PCB design.  You can see the entire Altium Academy channel here.  In the featured video, tech consultant Zach Peterson dives deep into the topic of copper pour and stitching, specifically what designers should and shouldn't do.  He explores via stitching, via fencing, radio frequency interference, return paths, and more.

To Pour or Not To Pour | Copper Pour in PCB Design, with Zach Peterson

Zach's video references another video, featuring Eric Bogatin, shown below. In this video Eric shows results of a simple 3D-EM analysis of two adjacent lines, with and without copper pour and stitching.  By looking at a linear graph, he concludes that a stitched pour does not help very much.  He stopped the analysis when he thought the result was good enough for digital traces. 

"You must Unlearn what You have Learned" by Eric Bogatin

However, Eric is not considering what happens when RF signals need to be isolated.  Instead of looking at transients, the analysis should look at S-parameters of the network to examine RF coupling in both directions (as you know, couplers are often directional) as well as line impedances which should be tuned to 50 ohms.  The analysis can be continued by putting in a lot more vias to create grounded coplanar waveguide lines, then model some absorber over the traces, then look at a channelized lid.  With some combination of all of these features, that cross-talk will be cut down to nearly zero (or 80 dB!) 

Once you understand the problem and all of the solutions, you will be in that sweet spot of being a PCB designer with microwave engineering chops or a microwave engineer with PCB design chops.  Both PCB designers and microwave engineers can learn a lot from each other, but it starts by admitting that you don't know everything.