[SI-LIST] Re: Validation of XTK results for clock skews
- From: Suchitha.V@xxxxxxxxxx
- To: ariazi@xxxxxxxxxxxxxxx
- Date: Tue, 10 Jul 2001 13:07:57 +0530
Sir,
Thankyou for your reply. It was indeed very helpful.
Thankyou for the rules that you have provided.
In your previous reply to the statement,
**********Because the minimum net segment length that XNS/TLC
can simulate and the time step are related by:
min_length =3D velocity * timestep***************
1) I am not clear what 3D means. Can you please explain this?
2) In regard to GTL simulation, you have said that a time step = 3D 0.005 ns
has been used frequently at Serverworks.
Do you control the minimum segment length(physical etch length) while
actually routing the GTL signals on your board?
3) *****************When analysis of overshoot/undershoot is of concern, the
following rule may be used:
1. Determine the signal rise time and take time step as 10% of rise
time.******************
To calculate the time step as 10% of rise time, can i use the rise time
given in the buffer model?
looking forward to your reply.
Thanks & Regards
Suchitha
From: abe riazi <ariazi@xxxxxxxxxxxxxxx>
Subject: [SI-LIST] Re: Validation of XTK results for clock skews
Date: Tue, 26 Jun 2001 10:34:10 -0700
Suchitha Wrote:
Thankyou for your reply. It was indeed very helpful.
I have one more question regarding time step.
**********Because the minimum net segment length that XNS/TLC
can simulate and the time step are related by:
min_length =3D velocity * timestep***************
For example, it time step=3D0.1ns(default) and velocity is 2.0ns/ft =
(Default),
then min_length considered for simulation is
min_length=3D2x0.1=3D0.2/12 =3D0.016". Am i correct?
The time step affects the accuracy of crosstalk. For purpose of =
measuring flight
time(ns delay), i have calculated the time step from the equation:
time_step =3D velocity x segment length/12 .
In case of measuring delay for high speed nets such as AGTL+ or GTL+ =
routed on
say 4 internal layers, does the time step have to be calculated =
considering the
minimum segment in each of the layers individually and then simulate , =
or can i
consider, the least minimum segment present from nets in all the 4 =
internal
layers and run the simulation? Will this change the timings =
considerably?
Hi Suchitha:
When using the equation:
min_length =3D velocity * timestep
we have to be careful about units and the difference between velocity =
and its reciprocal propagation delay. Therefore, when propagation =
delay equals 2ns/ft, the velocity equals 0.5 ft/ns =3D 6 inches /ns; =
If the time step is 0.1 ns as you had suggested then:
min_length =3D velocity x timestep =3D 6inches/ns x 0.1 ns =3D 0.6 =
inches.
At this point, let me share some rules that I have found useful in =
determining the optimum time step (and perhaps the experts on the list =
can suggest alternatives).
=20
When the objective of simulation is to investigate small delays (flight =
times), a satisfactory time step can be ascertained as follows:
1. Determine shortest trace/stub in the net.
2. convert it to time (by multiplying it with propagation delay).
3. Take time step less than 20% of that time.
4. Take display step same as time step.
When analysis of overshoot/undershoot is of concern, the following rule =
may be used:
1. Determine the signal rise time and take time step as 10% of rise =
time.
In regard to simulation of GTL, I and my signal integrity colleagues at =
Serverworks, have frequently used time step =3D 0.005 nS.
Furthermore, as Vinu Arumugham pointed out yesterday XTK can print =
warning messages suggesting appropriate time step settings. For =
instance, WARNING 92 (Time Step too large for coupled segment....), =
WARNING 602 (Time step < 0.0, set to 0.01), or WARNING 603 (Time step is =
suspiciously big) appear when time step requires optimization.
It should be also taken into consideration that while simulation =
accuracy can be enhanced by use of a very small time step, the =
simulation speed will be reduced as time step is decreased.=20
I hope you find the formulas and the rules that have been discussed =
valuable.
Best Regards,
Abe Riazi
ServerWorks
=20
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List archives are viewable at: //www.freelists.org/archives/si-list
Old list archives are viewable at: http://www.qsl.net/wb6tpu
Other related posts:
