I hope someone can shed some light in this, I am laying out a Ethernet to PI zero HAT and will be using the enc28j60 controller from microchip. People have reported getting 2.8mbps up and down speed on average with cheap modules form china with same chip.

My question is will impedance matching and differential parts make a big difference on traces 14.20 mm in length?

  • Copper 0.035mm T
  • Trace width 15 mil
  • trace <> trace space 4 mil
  • FR 4 material
  • 1.55mm thick
  • 2 Layer

When I do the impedance matching and calculations this will give me 107 Ohm Differential.

Or will I be better off just running normal routes and adding O ohm resistors in-line just in case there is a problem?
Example of matched traces enter image description here

Example of unmatched traces enter image description here


At 10 Mbps (which is, if I recall correctly, the only line rate that those chips support) you don't even have to bother with getting the impedance right due to the length of the traces. Just route it differentially and don't bother with the length matching. And don't add resistors, that's not how you fix a bad impedance match - for that you need either a re-spin with fixed dimensions or a matching network.

  • \$\begingroup\$ Hi Alex Thank you for the feedback. I was not trying to fix the impedance with the resistors that was for in case the signal is to strong. As you mentioned the chips can run at 10 Mbps but most people report 2.8 average. i would like to do what I can in the hardware to avoid any limitation. it seems like the limitation is the SPI software side of things but cant hurt to do the lenght matching anyway ? \$\endgroup\$ – Rustie0125 Sep 24 '17 at 6:10
  • \$\begingroup\$ @JohannvanNiekerk I have seen many ENC28J60-based circuits (and in general some 10 Mbps ETH systems) that were routed with much longer traces, in a very bad way (poor differential routing, poor length matching). They worked. Anyway the 2.8 Mbps limitation is likely due to SPI and software overhead. \$\endgroup\$ – next-hack Sep 24 '17 at 7:43
  • \$\begingroup\$ The raw line rate is always 10 Mbps, whether or not you can utilize that without large gaps between packets is a different story. Since the packets will actually be sent at 10 Mbps, that's the figure you want to use. Also, I think 10 Mbps Ethernet uses Machester encoding, so you need at least 10 MHz bandwidth instead of the 5 MHz you would need for 'raw' 10 Mbps. But at 10 MHz bandwidth, the traces and routing really make no appreciable difference. \$\endgroup\$ – alex.forencich Sep 24 '17 at 23:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.