A few questions about the application of ethernet controller and PHY chip and SFP module

Question

I have a question about the connection/application between the ethernet controller and PHY/SFP modules. Actually, there're so many scenarios for building up a NIC card, which makes me confused. For example, we can use an ethernet controller with the following connections:

1. controller ==> PHY chip ==> RJ45 connector ==> copper cable

2. controller ==> PHY chip ==> SFP module ==> fiber cable

3. controller ==> SFP module without PHY ==> fiber cable

4. controller ==> SFP module with PHY ==>copper cable

   ....

And the controller is the same, with the same serdes lanes. I don't quite understand why there are so many combinations to built up an ethernet link. Can someone give me some hints about the history of these setups? I can get the idea that the controller is working as MAC layer and PHY chip as physical layer, but the SFP thing really confuses me.

Another question is that, if the controller's serdes directly connected to SFP module (without PHY on the module), which protocol/interface will be used between controller and SFP module? If the SFP module only does the electrical-optical conversion, what data format is transmitted out? Since there is no PHY chip to deal with the physical layer stuff, can this communication be considered as data link layer communication?

About the controller, since it has the ability to directly interact with the SFP module, why the PHY chip vendor still need to implement the function to interface with SFP module?

Answer 1

The problem is many micro-controllers don't have transceivers built in, which require impedance control (and most likely add to the cost of IC development). The other problem is clocking, which low speed processors do not have clocks that support SERDES or transcievers. It really is determined by what interface the processor can support.

Another question is that, if the controller's serdes directly connected to SFP module (without PHY on the module), which protocol/interface will be used between controller and SFP module?

The biggest determining factor is what Media independent interface the processor can support (MII, RMII (50MHz), GMII (125MHz), or SGMII (serdes w/ trancievers)) Find what the processor can support, then that will determine what interface needs to be used.

An installation like this is usually when the microprocessor can't support A GMII or faster interface, and the phy translates between the microprocessor and the controller. Marvell makes some PHY's that can do that kind of translation.

controller ==> PHY chip ==> SFP module ==> fiber cable

About the controller, since it has the ability to directly interact with the SFP module, why the PHY chip vendor still need to implement the function to interface with SFP module?

No, that would be a waste, if you can do (for example) GMII on both the microprocessor and SFP, then no PHY is needed.