# Marvell PHY to KS8999 PHY connection

I'm trying to design a custom board with Marvell 88E1111 PHY and I want to connect this board to another board with KS8999 PHY on it. I want to connect two phy sides with capacitive coupling.

The distance between two boards is less than 5 inches and I have no option for using another PHY chip or using a transformer. I have searched through web and found these links:

Transformerless design

So I think that capacitive coupling is an option.

Micrel KS8999 supports capacitive coupling [here], but there is no specific data in 88e1111 datasheet about capacitive coupling.

Now my question is:

1. Is it possible to connect two phy sides?

2. If yes, how to connect lanes in terms of using series capacitors, pull-up voltage at 88E1111 side and pull-up resistor values?

KS8999 Datasheet

Is it possible to connect two phy sides?

Yes, it is poosible. It is less unpredicatable in 100BASE-TX (100 Mbps) mode than in 1000BASE-T (1 Gbps) mode, therefore you have a good chance. But it always case dependent: there is no single, universal solution.

Thus, prepare your design to a trial-and-error process by adding four 0603 places on each line in each pair on both sides of the de-coupling capacitor, like this:

    vcc1          vcc2
|             |
[ ]0603       [ ]0603
|             |
<>----o----|cap|----o----<>
|             |
[ ]0603       [ ]0603
|             |
gnd           gnd


From my own experience, i could say nothing about KS8999, but KS8995MA to KS8995MA worked normally in a transformer-less scenario with external termination resistors 49.9-ohm each pulled up to 3A3 and 0.1-uF capacitor array.

In a Gigabit scenario, capacitor arrays as decoupling work bad because of large parasitic cross-capacitance, discrete capacitors are needed here and (imo) preferable in any speed scenario.

Also, i can say nothing about the Marvell chip you mentioned, but some other Marvell chip (Gigabit switch) behaved unstable in my transformer-less scenario, therefore i recommend you to contact Marvell support (i think if you use 88E1111 you should have an NDA with Marvell that grants initial access to support) and try to use such connection method only if they say "explicit yes" on the transformer-less question. Otherwise, use a transformer -- it is cheap comparing with your debugging days.