Re: [SI-LIST] : Terminations scheme for bi-directional bus

Dennis Tomlinson (det@tellabs.com)
Tue, 30 Jun 1998 16:28:30 -0500

--------------575951273FFD8242930C28B5
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

Hi All,

George Harris wrote:

> Hi Susan,
>
> Your note has a lot of information so I'll intersperse my answers
> with your text.
>
> ----------
> From: M. Susan Tweeton[SMTP:mstweeto@collins.rockwell.com]
> Sent: Tuesday, June 30, 1998 12:43 PM
> To: si-list@silab.Eng.Sun.COM
> Subject: Re: [SI-LIST] : Terminations scheme for
> bi-directional bus
>
> All, sorry I haven't responded sooner, I've been out of
> town.
> In response to some of the questions and suggestions.
> First of all of these measurements are in SIMULATION.
>
> Have these boards been sent out for etch, or built, or otherwise
> committed? The best solution would be re-layout to place the
> transceivers next to the backplane connectors.

I'll second that. There's a discontinuity at the junction between daughter
card and backplane (for mid span cards) - with the backplane looking
at best, like 30 Ohms. The round trip time from driver to this discontinuity
and back to driver is almost one rise/fall time. This distorts transitions
even before launching onto the backplane and having to contend with all the
other ghoulies.

Again, if it's not too late, could you consider using one of the low voltage
bus interface logic families. In particular, GTL swings approximately
1.2V peak-peak, and slews slower than AC, ACT, or ACTQ by
a factor of 3 or so.
GTL would allow shunt termination on the backplane at both
extreme ends of your bus. This would do wonders for signal
beautification, and cut EMI emissions from your bus by a factor
of about 17.4 (-12.4 dB) due to the lowered voltage swing, and
decreased spectral bandwidth would decrease emissions further still.

Another caveat of using AC, ACT, or ACTQ as a bus interface
are those clamp diodes you'll find on the inputs. If your
system has a requirement for "hot card insertion", or individually
fused power on each daughter card, you are susceptible to a single
point failure causing an entire system to go down.
(Note, you may have a redundancy scheme to cover such things,
but it's a concern none-the-less).

Cheers, and good luck,

Dennis

>
>
> My system isn't
> in operation, and even if it were operating properly, I'd still
> have the
> same concerns. The biggest obstacle is the varied loading.
> There can
> be as few as 4 transceivers and as many as 8 transceivers on the
> bus at
> once.
>
> First incidence switching isn't necessary. The overshoot
> shouldn't be
> much of a problem either. It's about 5.93V and can always be
> clamped.
> I'm investigating the timing budget, but the signals are ugly,
> and I'd
> like to have them as robust as possible. Timing on this
> particular
> build may operate with these signals, but the first time we get
> an IC
> die shrink, we're in trouble. The thing I'm most concerned with
> is the
> ringback and EMI. The ringback crosses or comes near to crossing
> the
> threshold. Even if the timing is robust enough to accommodate
> the
> ringback, what's the EMI going to look like? Keep in mind this
> is an
> avionics system which has an extremely rigid EMI requirement.
> ^
> Now I am worried...
>
>
> When you refer to "terminating the backplane bus at each end of
> the bus"
> (ref Giovanni DiBenedetto & George Harris) are you referring to
> placing
> terminators on the backplane itself, physically at each end of
> the
> board? Or on the transceiver boards?
>
> In view of the 3 inch stubs, terminations would be more effective
> on the boards next to transceivers, but what if the end boards are not
> used? If configuration rules are acceptable, then end boards should have
> the terminations, but that would make them different from the middle
> boards, which is probably unacceptable.
>
> For now I have a 10 Ohm series resistors on each transceiver
> which looks
> the best so far. I'm looking into using schottky diode
> terminations
> like Roland Portman and Paul Thompson suggested.
>
> I don't think Schottky terminations will work here because of the
> distributed loads -- they work best for point-to-point signals, placed
> near the receiver.
>
> How about ferrites?
> Any ferrite experts out there?
>
> I never used ferrites in applications of this kind.
>
> How about some more discussion on placing the series terminators
> close
> to the transceiver vs placing them next to the connector to the
> backplane. As Larry Smith said, placing them next to the
> connectors
> will reduce the loading effects on the driver.
> Yet, others are adamant
> that they should be close to the driver. In simulation, there
> isn't a
> whole lot of difference. It is slightly (but not significantly)
> better
> with the resistor near the driver.
> For series terminations to be effective they have to be close to
> the driver. You didn't see a lot of difference because 10 ohms is not
> enough to make much difference.
>
> Parting thought:
>
> Simulation should be done with the driver in all possible positions with
> all possible load combinations. My guess is that the worst case will be
> full load with the driver in next to the last slot.
>
> Slow and fast silicon should also be simulated, not only typical.
>
> Thanks for all the responses so far.
>
> Susan
>
> Rockwell-Collins
> **** To unsubscribe from si-list: send e-mail to
> majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE
> si-list, for more more help, put HELP. si-list archives are accessible
> at http://www.qsl.net/wb6tpu/si-list ****
>
> ------------------------------------------------------------------------
>
> Part 1.2 Type: application/ms-tnef
> Encoding: base64

--------------575951273FFD8242930C28B5
Content-Type: text/html; charset=us-ascii
Content-Transfer-Encoding: 7bit

Hi All,

George Harris wrote:

Hi Susan,

Your note has a lot of information so I'll intersperse my answers
with your text.

        ----------
        From:   M. Susan Tweeton[SMTP:mstweeto@collins.rockwell.com]
        Sent:   Tuesday, June 30, 1998 12:43 PM
        To:     si-list@silab.Eng.Sun.COM
        Subject:        Re: [SI-LIST] : Terminations scheme for
bi-directional bus

        All,  sorry I haven't responded sooner, I've been out of
        town.
        In response to some of the questions and suggestions.
        First of all of these measurements are in SIMULATION.

        Have these boards been sent out for etch, or built, or otherwise
committed? The best solution would be re-layout to place the
transceivers next to the backplane connectors.

I'll second that.  There's a discontinuity at the junction between daughter
card and backplane (for mid span cards) - with the backplane looking
at best, like 30 Ohms. The round trip time from driver to this discontinuity
and back to driver is almost one rise/fall time. This distorts  transitions
even before launching onto the backplane and having to contend with all the
other ghoulies.

Again, if it's not too late, could you consider using one of the low voltage
bus interface logic families. In particular, GTL swings approximately
1.2V peak-peak, and slews slower than AC, ACT, or ACTQ by
a factor of 3 or so.
GTL would allow shunt termination on the backplane at both
extreme ends of your bus. This would do wonders for signal
beautification, and cut EMI emissions from your bus by a factor
of about 17.4 (-12.4 dB) due to the lowered voltage swing, and
decreased spectral bandwidth would decrease emissions further still.

Another caveat of using AC, ACT, or ACTQ as a bus interface
are those  clamp diodes you'll find on the inputs. If your
system has a requirement for "hot card insertion", or individually
fused power on each daughter card,  you are susceptible to a single
point failure causing an entire system to go down.
(Note, you may have a redundancy scheme to cover such things,
but it's a concern none-the-less).

Cheers, and good luck,

Dennis
 

 

         My system isn't
        in operation, and even if it were operating properly, I'd still
have the
        same concerns.  The biggest obstacle is the varied loading.
There can
        be as few as 4 transceivers and as many as 8 transceivers on the
bus at
        once.

        First incidence switching isn't necessary.  The overshoot
shouldn't be
        much of a problem either.  It's about 5.93V and can always be
clamped.
        I'm investigating the timing budget, but the signals are ugly,
and I'd
        like to have them as robust as possible.  Timing on this
particular
        build may operate with these signals, but the first time we get
an IC
        die shrink, we're in trouble.  The thing I'm most concerned with
is the
        ringback and EMI. The ringback crosses or comes near to crossing
the
        threshold.  Even if the timing is robust enough to accommodate
the
        ringback, what's the EMI going to look like?  Keep in mind this
is an
        avionics system which has an extremely rigid EMI requirement.
    ^
        Now I am worried...
 

        When you refer to "terminating the backplane bus at each end of
the bus"
        (ref Giovanni DiBenedetto & George Harris) are you referring to
placing
        terminators on the backplane itself, physically at each end of
the
        board?   Or on the transceiver boards?

In view of the 3 inch stubs, terminations would be more effective
on the boards next to transceivers, but what if the end boards are not
used? If configuration rules are acceptable, then end boards should have
the terminations, but that would make them different from the middle
boards, which is probably unacceptable.

        For now I have a 10 Ohm series resistors on each transceiver
which looks
        the best so far.  I'm looking into using schottky diode
terminations
        like Roland Portman and Paul Thompson suggested.

I don't think Schottky terminations will work here because of the
distributed loads -- they work best for point-to-point signals, placed
near the receiver.

          How about ferrites?
        Any ferrite experts out there?

I never used ferrites in applications of this kind.

        How about some more discussion on placing the series terminators
close
        to the transceiver vs placing them next to the connector to the
        backplane.  As Larry Smith said, placing them next to the
connectors
        will reduce the loading effects on the driver.
         Yet, others are adamant
        that they should be close to the driver.  In simulation, there
isn't a
        whole lot of difference.  It is slightly (but not significantly)
better
        with the resistor near the driver.
        For series terminations to be effective they have to be close to
the driver. You didn't see a lot of difference because 10 ohms is not
enough to make much difference.

Parting thought:

Simulation should be done with the driver in all possible positions with
all possible load combinations. My guess is that the worst case will be
full load with the driver in next to the last slot.

Slow and fast silicon should also be simulated, not only typical.

        Thanks for all the responses so far.

        Susan

        Rockwell-Collins
        **** To unsubscribe from si-list: send e-mail to
majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE
si-list, for more more help, put HELP.  si-list archives are accessible
at http://www.qsl.net/wb6tpu/si-list ****

  ------------------------------------------------------------------------

   Part 1.2       Type: application/ms-tnef
              Encoding: base64

  --------------575951273FFD8242930C28B5-- **** To unsubscribe from si-list: send e-mail to majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE si-list, for more more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****