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

Re: Single ended XGMII




Curt and all,

Could I suggest to reduce the driving capability of the
output buffers to 4 mA for data and control lines (36 lines
total) and keep 8 mA only for the recovered clock ? I used it
the past for Fiber Channel chips and it was very effective.

Essentially, it reduces the switching noise by half.

Jaime

Curt Berg wrote:

> A few comments on using a single ended XGMII:
>
> - Today we have several Quad SERDES chips in production, operating
>   at 1.25/2.5 Gbit/s with 8B or 10B interfaces. Most
>   use a 3.3V I/O swing, and I would guess about 40 ohm (nominal)
>   LVTTL/LVCMOS drivers. This effectively is a 32 bit interface.
>   I would say, these devices seem to function well for many companies.
> - Some vendors are event moving to higher integration then
>   Quads, still using single ended interface. This further indicates
>   that noise is not a major problem.
> - Single ended interface on the MAC side is not an issue. There
>   are many ASIC in production today with 400+ single ended I/Os,
>   and board with several tens of thousands of single ended nets.
> - If we for XGMII use a (1.5 V swing) HSTL instead, and
>   move to a 50-55 ohm impedance controlled driver, we should be
>   able reduce the effective noise levels by close to a factor
>   of three.
>
> My conclusion is that the industry keeps pushing the
> capabilities of single ended interfaces, just because
> differential interfaces double this pincount.
>
> If we can reduce the noise by a factor of 3, and not
> move up much in frequency, from existing proven solutions,
> I believe we have a workable solution.
>
> -Curt Berg-