As we know, Ethernet PHY needs to be connected to external devices through Ethernet transformer, RJ45 interface. The interface between PHY and Ethernet transformer is called MDI interface, which is media-related interface (which is relative to MII). In the 100-megabit Ethernet mode, MDI is two pairs of differential lines, and in the Gigabit mode, it is four pairs of differential lines. Data sheet or application manual of some PHY chips will give MII/RMM/GMII/RGMII interface and MDI interface equal length rule, but few manufacturers mention the difference equal length rule between Ethernet transformer and RJ45.
In the early product design, there was no need for special attention. The difference between RJ45 and Ethernet transformer was similar to each other in length, but in recent product design, the need for anti-surge design of Ethernet led to a lot of difference in length. Because of the uncertainty of this part of the equal length rule and the lack of relevant information for reference, I have tried two cases of equal length and unequal length.
Differential Equal Length of RJ45-Ethernet Transformer
If the difference on the left is not serpentine, the length of the two difference lines will be much different. Therefore, in order to avoid problems, I intentionally make the length of two pairs of differential lines consistent. Careful readers may have found that there are holes in the route, which is one of the major taboos of Gigabit Ethernet PCB routing, but in the face of mounted RJ45 interface, this is the only way.
RJ45-Ethernet Transformer Differential to Unequal Length
Looking at the figure below, it is easy to see that the four pairs of differential lines between the Ethernet transformer and RJ45 must be unequal in length. Of course, the two lines in the differential pair are treated equally.
The above two cases are designed by myself, using iperf to test the throughput of Ethernet (I have no condition to use SmartBits). The results are as follows:
It is conceivable that, if not affected by CPU processing capacity, the difference between RJ45-Ethernet transformers does not affect the actual throughput. Of course, from this table, we can clearly see that PowerPC has super network processing power, 36.5% of the CPU occupancy only uses one core, the other core is completely idle.
Today, I looked at it in the forum. The more reliable statement is as follows:
These four pairs of MDI signals do not need equal length processing. To be safe, you need to confirm whether your PHY has automatic alignment of data between differential pairs and how deep the FIFO of PHY is.
Now it can be basically determined that the difference between RJ45 and Ethernet transformer does not require equal length processing.