Routing Differential Pairs - 90 ohm (USB) and 100 ohm (Ethernet) - On 2 Layer Board (1.6mm thick)

Thread Starter

Mahonroy

Joined Oct 21, 2014
332
I have a design that has onboard ethernet and USB (CM3 with a LAN9512 chip, and FSUSB42MUX chip). I need to rout the USB lines as a differential pair at 90 ohm impedance, and the ethernet lines at 100 ohm impedance differential pair.

On my previous design, I miscalculated the trace width and spacing, but lucked out and it still worked most of the time. We would sometimes have to reset a device in order for the USB to connect, or for the ethernet to connect. I have no idea if this is related to this or not. I have read that if the traces are less than 70mm, that the impedance doesn't really matter at this speed of ethernet and USB 2.0 - technically it matters but it shouldn't affect the signal enough to make it matter. All of my differential pair traces were under 50mm on this board.

My miscalculations had me routing between 8-10 mil wide traces, with a 8-10 mil gap between the pairs, and 25 mil spacing to the ground pour and other signals. I length matched all pairs to within 1mm. This yielded like 150 ohm impedance, not 90 or 100. According to online calculators and Saturn PCB Design tool, I should have had a trace like 35 mil wide, and 5 mil gap between the two. So I was way off.... but like I mentioned above it still mostly worked. Once it connected it seemed to stay connected.

So with that said... would it work better for me to enlarge the traces as much as I can get away with? There is no way I can rout 35 mil differential pair traces for this, that's huge. I can make 16 mil traces work though. And remember, my traces are around 60mm long on this design (still under the magic 70mm number). Is 16 better than 10 in this application? Or if I cannot meet the 35 mil trace specifications then it doesn't matter and I can run whatever width I want? I wonder if this small change will improve the reliability of both the ethernet and USB, or if that wouldn't matter.

Part of the challenge with this is its a standard 2 layer board, 1.6mm thick. It looks like people tend to do a 4 layer board, keeping the signal and ground layers close together, which allows them to run skinnier traces for differential pairs. I can't do this unfortunately.

Any help or advice on this is greatly appreciated, thanks!
 
Top