How to connect the PHY crystal for RMII?

Question

I'm attempting to implement Ethernet communication for the first time and have run into an issue with my PHY. I'm using the KSZ8091RNBCA. According to the datasheet, when running in RMII mode, the XO pin is supposed to be left unconnected (Tables 2-3, 3-2). Figure 3-4, shown below, shows that my external 50MHz crystal needs to be connected to both the XI pin on the PHY and the REF_CLK line.

Since both paths are shown coming out of the crystal, does REF_CLK come off the same pin as XI, or does it come off of the other pin, where the XO would normally go?

^{simulate this circuit – Schematic created using CircuitLab}

Answer 1

Figure 3-4 shows an oscillator, not a crystal.
Similarly, the datasheet says "RMII - 50 MHZ CLOCK MODE [...] An external 50 MHz clock source (oscillator) connected to XI"

A "crystal" would be excited by XO and the desired clock frequency would come into XI. In this context, an "oscillator" is something which excites itself externally, and just provides a clock to your device's XI or REF_CLK input.

The datasheet is trying to tell you to use some external device which provides a 50MHz clock without needing to be excited by XO.

Answer 2

Figure 3-4 is showing an oscillator instead of a crystal. The XO pin on the PHY is meant to provide an excitation source for a crystal located close to the IC. However, an oscillator contains this excitation source along with other circuitry to stabilize it and provide an output clock. Connect this output to both the PHY's XI pin and the MAC's REF_CLK pin. It is recommended to include an in-line resistor between the oscillator and the PHY/MAC.

This is addressed a little bit more clearly in the Schematic Checklist for the LAN8710A. Numbers were added to the image below to match the circuits in the question. While the two connections in this context are shown for RMII and MII modes, respectively, it highlights the difference between an oscillator and a crystal. An oscillator contains its own excitation source but the crystal must be driven by an external excitation source.

Adding an in-line resistor between the oscillator and both the PHY and the MAC allows you to adjust for impedance mismatches. The physical layout will impact the impedance of the traces. On many PCBAs, the length of the trace will not be an issue. Also, the width of the trace from oscillator to either resistor can have an impact and may cause instability. Not that I've had that issue before...