2

I need to communicate between a microcontroller that has an integrated MAC (Datasheet) and a microprocessor that has an integrated MAC and PHY (Datasheet). The pins of the MAC on the processor are already used by design, so I use an Ethernet transceiver with an integrated PHY between the two (Datasheet). Here is the schematics :

enter image description here

The chips are on the same board. There will not be a RJ45 connector. The maximum distance between them will be some centimeters. I do not know yet if I will use 10 or 100 Mbps.

My questions are as follows:

  • Do "magnetics" and "transformer" are the same thing?
  • Do I need one? If no, in which cases do I need one?
  • Do I need to use impedance matching for the Ethernet wires, even for such a small distance?

Thanks.

uz3il
  • 81
  • 5
  • 1
    You can't call it ethernet without the magnetics; it becomes something else in name. Yes, you need impedance matching. – Andy aka Jan 18 '23 at 09:28
  • 1
    yes, in the context of Ethernet, "Magnetics == transformers" – Marcus Müller Jan 18 '23 at 09:36
  • 1
    yes, impedance matching makes sense for any digital bus (100 Mb/s ethernet has not a negligible bandwidth – though it probably really won't matter over a few cm). Though you'll probably see it more as "termination" in this case, as the trace impedance at short lengths only plays a negligible role compared to discrete termination components for the impedance seen by a source. – Marcus Müller Jan 18 '23 at 09:36
  • 1
    Regarding Olin Lanthrop's answer in the linked question: Yes, your DP8348 is one of these PHYs that needs a transformer to implement the bipolar driving of the line, seeing that the reference circuit connects the transformer's middle tap to Vdd. So, you will need a transformer. At least 1 per direction – you might get away without a second one. So, I think this will be [1,14€](https://www.digikey.de/short/7nwrntjh) in your BOM that you cannot, with reasonable development effort, avoid. Maybe with a lot of lab testing and a different kind of PHY than the DP8348, but not with this one. – Marcus Müller Jan 18 '23 at 09:59
  • Thanks for your answer ! How could I get away without a second one? Does it concern the RMII interface? Unfortunately, BOM cost is a real concern... I think I might look at 10BASE-T to SPI Ethernet transceivers as well, as I have an SPI interface on the MCU's side. – uz3il Jan 18 '23 at 10:33
  • If BOM cost is a concern, drop the DP8348 for a slightly more expensive PHY maybe, which doesn't need the center tap. Then, test with AC coupling through capacitors instead of through a transformer, and fiddle with the board design and termination until you get it to work (or prove you cannot get it to work). I'd honestly not recommend this; this feels like trying to shave off a few cents off the BOM that will come back and haunt you through unreliable operation of your mass-produced products at the many customers. (if this is not mass-produced, you wouldn't care about the BOM cost this much) – Marcus Müller Jan 18 '23 at 13:52

1 Answers1

1

  1. Yes in Ethernet designs magnetics mean the transformer.

  2. You need to check from the data sheets, application notes or manufacturer support if both devices support capacitive coupling or not. If they don't support or it is unknown, then you need a transformer. If they do support then you don't need a transformer. Usually you can't use 10Mbps via capacitive coupling so you are better off with 100Mbps.

  3. If you want a properly working product it makes sense to route the wires properly as impedance controlled differential pairs. You likely already have such a complex design that you already use impedance control.

Justme
  • 127,425
  • 3
  • 97
  • 261
  • And if you do need a transformer, it's only needed for biasing; for which, very simple parts are available; you don't need the full kit. Example: Bourns SM453229-381N7Y; Abracon ALANC10001-CDB3DT; TDK ALT4532-201-T001; etc. – Tim Williams Jan 18 '23 at 15:23