AM62L: Simulation doesn't match the reality

We searched for a bit, and did find that in the IBIS model the RGMII interfaces are all labled as IO and not as LVCMOS. We changed it, and got better results, even tho with a lot of reflections.

Hello pjytecBSC,

Thank you for the inputs.

Can you please share a waveform with the reconfigured IO as i check internally.

Regards,

Sreenivasa

This is what it looks like right now.

The timing is at least somewhere near our measurement, but still a bit off. But remarkable for me is the reflections on that, even tho the ramp is much slower.

This is the simulation for CLK/RGMII

And this is the measurement

Hello pjytecBSC,

Thank you.

Please refer the inputs i received from the simulation expert: Could you please elaborate the concern.

Can they clarify what is their exact concern here?

The AM62L IO buffers should not be changed by the customer so I am not clear what exactly they mean by this. Whatever is in the IBIS model is what they must use for simulations.

We need some clarity on what is their concern, in order to provide appropriate responses.

Are you seeing any performance issue with the board?

Regards,

Sreenivasa

Hi.

No, as of now, I don't have any performance issues (Climate Test is still missing). But as we are developing a SoM which can be used with other baseboards, it is crucial that we ensure a proper functionality of the interface.

The pictures I did show you is "only" the Phy on the SoM which is about 1.5 inches away from the SoC. The one on the baseboard looks even worse. And to provide a customer with an easy to integrate interface, it should also be according to the IEEE specification or at least to the datasheet of the used phy (Ti DP83867IRRGZ). For that, we need a rise/fall time of 0.75 ns, where we are currently not even close to. (Sidenote: And the strange behaviour of the Phy is not helping with that either, neither the allegedly missing/hidden option to set the driver strength in the SoC).

Of course, I don't want to tinker with the IBIS model. But I hope we are on the same page, that it is kind of strange that when the RGMII is marked as LVCMOS in the datasheet it is not represented in this way in the IBIS model. Which sparks the question if the model is correct.

So my concerns as TL;DR: SoC/Phy combination doesn't provide the necessary rise/fall time to comply with IEEE, hence making the interface hardly sellable. Simulation is vastly different, causing question about the model.

Hello pjytecBSC,

Thank you.

Let me check with the expert and comeback.

Regarding the  buffer type for the RGMII please see the below picture.

Regards,

sreenivasa

Any update on this?

Hello phytecBSC,

below is the discussion i am having with the team.

Their response is not the most clear but I was able to gather two main points which I shall address below:

 0.75 ns rise/fall time: This might be stated in the RGMII standard spec but it is not a must-have from the PHY perspective. We have confirmed that with our PHY team for DP83867IRR functionality. The only care about is the setup/hold time. They do not care about r/f time as long as setup/hold times are met.

I am waiting for additional inputs.

Regards,

Sreenivasa

Well... I am not on the same page here, to be honest. But maybe I forgot to mention that we are speaking of 1000 Mbps and not 100 Mbps

So according to your datasheet, the rise fall time is at max 0.75 ns. You surely can add some inaccuracy because of the scope and all, but it should not be that far off.

And only then you would be able to fulfil some undocumented setup and hold time, which is not even marked in the timing diagram and is solely based on the skew time, which is a sufficient definition, as long as rise/fall time are in spec.

So, not quite sure what to do with the response of your team, when it is pretty much contradicting to the datasheet. But maybe the GbE was the missing information.

Hello phytecBSC,

Thank you.

I am checking with the expert if he has additional thoughts.

Regards,

Sreenivasa

Hello phytecBSC,

Is there a TI support/sales contact you are working with.

Regards,

Sreenivasa

Not yet

Hello phytecBSC,

Thank you.

Could you please send me a private message to continue the discussion.

Regards,

Sreenivasa