how fast can the FF be operated

[Neeraj Parik]

Hi,
Here is a hypothetical question. I have FF with 50ps of setup time,
10 ps of hold time; however this FF has 200 ps of clk - Q delay. I want
to use this FF to capture an 8Gbps input stream (125 ps). Is this
possible ( mainly due to CLK-Q delay) ? I am open to using advance
techniques like pipeline timing etc.

Neeraj

 

[John Larkin]

On Thu, 10 Mar 2005 13:55:50 -0800, Neeraj Parik
<neeraj.parik@xilinx.com> wrote:

Hi,
Here is a hypothetical question. I have FF with 50ps of setup time,
10 ps of hold time; however this FF has 200 ps of clk - Q delay. I want
to use this FF to capture an 8Gbps input stream (125 ps). Is this
possible ( mainly due to CLK-Q delay) ? I am open to using advance
techniques like pipeline timing etc.

Neeraj

Hi,

What part are you using?

The NBSG53A can clock to >8 GHz, so presumably propagates data at that
rate. Typical setup/hold is 10/7 ps. You might consider interleaving 2
or 4 of these to let their outputs settle long enough to pass the data
on to the rest of the system.

Tricky stuff, at this speed. These parts are so small it's hard to
keep the trace impedances matched as the traces neck down into the
chips. Think *very thin* dielectrics.

John

 

[Joerg]

Hello John,

Tricky stuff, at this speed. These parts are so small it's hard to
keep the trace impedances matched as the traces neck down into the
chips. Think *very thin* dielectrics.

Can't you via down to the next layer somewhere before the bottleneck and
then use a plane above and below instead of just below? The transition
may be a bit nasty but should be feasible.

Regards, Joerg

http://www.analogconsultants.com

 

[John Larkin]

On Fri, 11 Mar 2005 00:19:43 GMT, Joerg
<notthisjoergsch@removethispacbell.net> wrote:

Hello John,

Tricky stuff, at this speed. These parts are so small it's hard to
keep the trace impedances matched as the traces neck down into the
chips. Think *very thin* dielectrics.

Can't you via down to the next layer somewhere before the bottleneck and
then use a plane above and below instead of just below? The transition
may be a bit nasty but should be feasible.

Maybe so, but I'm not brave enough to use vias below about 200 ps. If
you TDR a matched line that hops through a via, it usually look like a
lumped capacitor. I guess it's possible to keep a via at the
characteristic impedance somehow, but we just try to keep all the fast
sigs on the top layer. For complex logic, that would be impossible,
but we've never done that.

We're just now doing a fiducial pulse generator that drives an e-o
modulator. It uses a GigaLogic gate to make a fast (40 ps) edge, then
differentiates it into the input of one of the Giga parts that has
variable output swing (we DAC that) and then through two GaAs
distributed amps. We're getting a clean 100 ps impulse variable from 0
to almost 7 volts, all on FR4 with a 10 mil first dielectric layer,
necking down the 50 ohm traces into the rediculously small chips. It
all gets hot.

John

 

[Joerg]

Hello John,

We're just now doing a fiducial pulse generator that drives an e-o
modulator. It uses a GigaLogic gate to make a fast (40 ps) edge, then
differentiates it into the input of one of the Giga parts that has
variable output swing (we DAC that) and then through two GaAs
distributed amps. We're getting a clean 100 ps impulse variable from 0
to almost 7 volts, all on FR4 with a 10 mil first dielectric layer,
necking down the 50 ohm traces into the rediculously small chips. It
all gets hot.

Wow. Is it Onsemi GigaComm? These are really expensive but if it's
high-end they are sure nice devices to have at hand. I rarely designed
at that kind of budget level so fast stuff had to be discrete. But it
wasn't this fast.

Regards, Joerg

http://www.analogconsultants.com

 

[John Larkin]

On Fri, 11 Mar 2005 17:59:04 GMT, Joerg
<notthisjoergsch@removethispacbell.net> wrote:

Hello John,

We're just now doing a fiducial pulse generator that drives an e-o
modulator. It uses a GigaLogic gate to make a fast (40 ps) edge, then
differentiates it into the input of one of the Giga parts that has
variable output swing (we DAC that) and then through two GaAs
distributed amps. We're getting a clean 100 ps impulse variable from 0
to almost 7 volts, all on FR4 with a 10 mil first dielectric layer,
necking down the 50 ohm traces into the rediculously small chips. It
all gets hot.

Wow. Is it Onsemi GigaComm?

Right.

These are really expensive

Ballpark $30 each, although some of the not-so-fast parts are cheaper.
EclipsLite gates are $5-ish and come in sensible packages.

but if it's
high-end they are sure nice devices to have at hand. I rarely designed
at that kind of budget level so fast stuff had to be discrete. But it
wasn't this fast.

It's amazing the parts you can buy these days. They are getting mighty
hard to see and solder, though.

John