RE: [SI-LIST] : Looking Inside IBIS Model

Beal, Weston ([email protected])
Tue, 18 May 1999 11:21:00 -0500

Steve,
Almost all the datasheets I've seen provide minimum and maximum times for
Tco. These should include all cases including rising and falling waveforms.
I agree that TIME_TO_VM can be significantly differenct for rising and
falling edges. That's why the TIME_TO_VM parameter in the QUAD model has
two different numbers for rising and falling. If you will calculate the two
TIME_TO_VM parameters (or let mtest do it), put the TIME_TO_VM in the QUAD
driver model, and enable VM correction in XNS, all your time deratings will
come out correct.
On the other hand, as I stated in a previous email, the timing test load
should be placed at the pin. Including the TIME_TO_VM parameter in the
model with mtest puts the timing test load at the die. To be more correct,
you might want to calculate TIME_TO_VM (rising and falling) to the pin and
do your manual subtraction with the rising and falling cases seperately.
In your example, if 2.5ns is the Tco_max then your Tflight either rising or
falling is 3.0ns and 3.1ns breaks your timing.
Regards,
Weston Beal

-----Original Message-----
From: [email protected] <mailto:[email protected]>
[mailto:[email protected]]
<mailto:[mailto:[email protected]]>
Sent: Monday, 17 May, 1999 8:33 PM
To: '[email protected]'
Subject: RE: [SI-LIST] : Looking Inside IBIS Model

Dear SI-List,
I'm sorry I sent something that most of you cannot read, here I post it
again (and made some modification) with pure text format:
Abe,
I agree with you.
I believe that the test loading value specified in IBIS should be consistent
with datasheet.
If we can get Tco [clock to output] from driver's spec and Tsetup (Thold)
[setup time, hold time] from receiver's spec ( and with clock's info.
available), we can determine the max. and min. flight time for the data.
To determine the max. flight time (assume zero skew, jitter), I'll use the
formula:
Tflight_max = Tcycle - Tco_max -Tsetup
And to determine the min. flight ti
Tflight_min = Thold - Tco_min

And in my opinion, I think the Tco can be decomposed as IC internal delay
plus TIME_TO_VM, which depends on loading outside the driver. To avoid
double subtraction with the TIME_TO_VM on the right side of above formula,
the flight time obtained from simulation (IDD file generated by Quad's tool
with VM correction DISABLED) should be corrected by subtracting the
TIME_TO_VM.
In many cases, where the TIME_TO_VM will be different for rising and
falling.
And sometimes, the asymmetry between rising and falling will be quite
significant, that will result in some persecution. Because we'll have only
one Tco in the spec.
For example, I have:
one driver A with Tco =2.5nS
one receiver B with Tsetup=2nS
running speed is 133MHz (Tcycle=7.5 nS).
TIME_TO_VM for the driver is 0.8nS on rising and 1.8nS on falling (1 nS
difference!).
According to above formula Tflight_max =7.5 - 2.5 - 2 = 3nS
What if my simulation result reports the Tflight_rise (rising) = 3.1nS and
Tflight_fall (falling) = 2.7nS for some signal from layout database? (Not
caused by ringback!)
It happened because the real loading in actual application won't necessarily
be the same as the test loading
used in datasheet. But how should I deal with it? If it is true that there
exist such asymmetry between rising and falling,
why doesn't the vendor privide more info on Tco?
Any comments will also be highly appreciated!
With best regards,

Steve Ting
EDA/CAE Engineer
INVENTEC CORPORATION

-----Original Message-----
From: Beal, Weston [ mailto:[email protected]
<mailto:[email protected]>
<mailto:[email protected] <mailto:[email protected]> > ]
Sent: Monday, May 17, 1999 10:03 PM
To: '[email protected]'
Subject: RE: [SI-LIST] : Looking Inside IBIS Model

Dear IBIS people,
This message got cut off before it reached me. Did anyone get the whole
thing?
Anyway, I think a correction is important here. Steve wrote, "To avoid
double subtraction with the TIME_TO_VM on the right side of above formula,
the flight time obtained from simulation (IDD file generated by Quad's tool)
should be corrected by subtracting the TIME_TO_VM." BE CAREFUL with this!
The TIME_TO_VM parameter should be included in the driver model and XNS
subtracts TIME_TO_VM from the simulated flight time by default. The time
given in the IDD file is the corrected time that you should use in your
timing margin calculation. It is important to understand what you actually
simulated and and calculated. The automatic TIME_TO_VM correction can be
disabled or the TIME_TO_VM parameter might be missing from the model. Check
these details! The difference between an ameteur and a professional is in
the details.
Regards,
Weston Beal

-----Original Message-----
From: [email protected] <mailto:[email protected]> [
mailto:[email protected] <mailto:[email protected]>
<mailto:[email protected]
<mailto:[email protected]> > ]
Sent: Monday, 17 May, 1999 6:07 AM
To: '[email protected]'
Subject: RE: [SI-LIST] : Looking Inside IBIS Model

Abe,
I agree with you.
I believe that t test loading value specified in IBIS should be consistent
with datasheet.
If we can get Tco [clock to output] from driver's spec and Tsetup (Thold)
[setup time, hold time] from receiver's spec ( and with clock's info.
available), we can determine the max. and min. flight time for the data.
To determine the max. flight time (assume zero skew, jitter), I'll use the
formula:
Tflight_max = Tcycle - Tco_max -Tsetup And to determine the min. flight
time, I'll use:
Tflight_min = Thold - Tco_min
And in my opinion, I think the Tco can be decomposed as IC
internal
delay plus
TIME_TO_VM, which depends on loading outside the driver. To avoid double
subtraction with the
TIME_TO_VM on the right side of above formula, the flight time obtained from
simulation (IDD file generated by Quad's tool) should be corrected by
subtracting the TIME_TO_VM.

In many cases, where the TIME_TO_VM will be different for rising and
falling.
And sometimes, the asymmetry between rising and falling will be quite
significant, that will result in some persecution. Because we'll have only
one Tco in the spec.
For example, I have
one driver A with Tco =2.5nS

**** To unsubscribe from si-list: send e-mail to [email protected]. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****