Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: XAUI AC coupling


OK, I asked for it... Now I'm forced to respond.

A PHY module to ASIC connection, with the ASIC being in-line and
eventually connected to a switch fabric, is very likely to be a XAUI
implementation. If the module is pluggable, then it is very likely that
the XAUI link would be AC-coupled to insure maximum interoperability.
However, if the module is fixed, there are non negligible cost,
reliability and performance advantages to employing DC-coupling instead
if applicable. There is no risk. My point all along, is that coupling is
an implementation detail and should not be a standard mandate.

Your PECL example reflects information an implementer should be able to
glean from a manufacturers data sheet. Many XAUI devices will be fixed.
The implementer simply reads the relevant data sheets for both XAUI
devices and determines whether or not AC-coupling is required. If
AC-coupling is not required, the implementer may choose to DC-couple.
The determination of signal coupling requirements is standard practice
for chip-to-chip interconnects.    

Your third paragraph seems to describe a scenario where an implementer
makes a bad decision to employ DC-coupling where the device specs for
the two XAUI devices employed in the link dictated AC-coupling. I was
unaware that the purpose of the standard was to force suboptimal
implementations in case an implementer misinterprets device data sheets.
I believe that most implementers would be incensed by such imposing
regulations. I certainly hope that the same implementer doesn't rely on
the standard for all other aspects of XAUI link implementation, such as
power supply decoupling, trace layout, connector choice, via design,
etc. to insure that their XAUI links work reliably.

I don't understand the relevance of LVDS to this discussion, please

I agree that if either an implementer is uncertain about DC bias or DC
bias itself is uncertain, that AC-coupling should be used. However, you
seem to be describing a scenario where too much uncertainty exists. It
is highly likely that the operation of the XAUI link will be uncertain
in this case.

To conclude, your assumed XAUI configuration is system ASIC to
transceiver module which exemplifies only one possible XAUI application
and one in which DC-coupling is applicable and preferred in many
instances. In addition your desire is to impose suboptimal
implementations on all XAUI links in case an implementer happens to make
a mistake. I have to respectfully disagree that either argument dictates
that XAUI AC-coupling is technically required.

Best Regards,

Vipul Bhatt wrote:
> Rich,
> It seems to me that AC coupling deserves more to be "required" than
> "not required".
> PHY module manufacturers will gain maximum economies of scale by
> being able to sell the same PHY module to multiple switch
> manufacturers, a scenario made possible by XAUI. But each party will
> control only one end of the link. Without AC coupling, that may
> become a risky scenario.
> Take PECL as an example. PECL gate inputs need resistors to set up
> DC bias and to match impedance. Some ICs implement this
> resistor/bias network internally, some don't. PECL outputs (the
> "hanging" emitters) also need a similar resistor network to set up
> DC bias and to match impedance. Often, a termination is a single
> resistor; often it is two resistors - one pulled high and one pulled
> low, to achieve the same (small signal) impedance but a different
> level of bias current.
> Now consider the COMBINATIONS of inputs and outputs, if DC coupled.
> If the transmit side uses a different VDD than that assumed by the
> receive side (or vice versa), the bias points at both ends of the
> link will change. When the AC signal is added, it will span a
> voltage range it wasn't ptimally designed to span. Even with the
> same VDD at both ends, the different ways of implementing resistor
> networks would change the assumed bias points. If we make AC
> coupling optional, and some use it, the bias point assumptions
> change again. It all gets more chaotic if you add LVDS option to
> this scenario.
> The safest way to eliminate DC bias uncertainties is to ensure AC
> coupling. I would find it risky to blindly design a DC coupled
> driver/receiver where one end of the link is under my control and
> the other end is not. In a market where you will see more than 15
> switch manufacturers and more than 20 PHY module manufacturers
> announcing products within a 12-month span, it is not realistic to
> assume that they will work it out easily. Therefore, I think we
> should specify AC coupling.
> Regards,
> Vipul
> vipul.bhatt@xxxxxxxxxxx
> (408)542-4113
Richard Taborek Sr.                 Phone: 408-845-6102       
Chief Technology Officer             Cell: 408-832-3957
nSerial Corporation                   Fax: 408-845-6114
2500-5 Augustine Dr.        mailto:rtaborek@xxxxxxxxxxx
Santa Clara, CA 95054