[SI-LIST] Re: Looking for an algorithm to clip waveforms
- From: Vishwanath Elechithaya <elechi@xxxxxxxxx>
- To: si-list@xxxxxxxxxxxxx
- Date: Tue, 1 Jun 2010 15:52:21 +0530
Hi Tim,
That's the basic IDEA.I am clipping off all the initial delay with the
waveforms, i.e, where I see significant changes in the waveforms. This gives
me an excellent correlation of spice to IBIS with MAX corner. However, the
fall edges of the waveforms do not match for the TYP & Min Corner (IBIS
waveforms were faster than the Spice). After a little more struggle, I added
little bit of delays on the fall waveforms of TYP & MIN, this gave me good
correlation of TYP & MIN waveforms too. Now the problem is that how to
arrive at the "initial delay" that NEEDS to be present on FALL waveforms so
as to match IBIS vs Spice!.
This is the approach I try to build.
1. Clip the waveforms (MIN, TYP & Max) to the maximum extent (till there is
significant change in the waveform).
2. Terminate the Waveform window to 1/freq (Ex: for a 1333MHz system, time
window of the VT table = 750ps).
3. Add "INITIAL DELAY" s to the typ & min Waveforms (but how much?) to match
the Spice waveforms.
There is a least preferred approach that to simulate the IBIS models &
extract the mismatch of the waveforms which can be used for calculating the
"INITIAL DELAY" that needs to be added. But would rather be an iterative
process.
Regards,
elechi
On Fri, May 28, 2010 at 10:04 PM, Timothy Coyle
<tim.coyle@xxxxxxxxxxxxxxxx>wrote:
> Hi Elechi,
> It will be difficult to come up with a general algorithm to handle over
> clocked VT table waveforms. But if you really wanted to you could read in a
> waveform and have it discard data until a significant change in slope
> occurs. This would cut out initial data before the waveform transitions.
>
> In theory it should be something that you don't have to worry about
> provided
> that your SPICE model cells do not have a long internal delay. You want
> your
> VT waveforms to settle out and that should be the time window that you
> generate your VT data under. But let's say that it takes 5ns for your
> signal
> to start transitioning from high to low. Then you may want to get rid of
> that initial data where the signal is not transitioning. Rather than an
> automatic algorithm you could have an entry field where a user could enter
> how much time to cut off. This is how I implemented this type of feature in
> a commerical IBIS modeling tool I develop.
>
> On a side note you need to make sure that you are not trying to line up the
> Tco delay of the SPICE model to the IBIS model because they do not have to
> match up at all. Remember IBIS does not do anything with internal timing
> such as Tco or prop delay. So the only time you should be concerned about
> the time window length of your VT waveforms is if they are longer than the
> 1/2 clock period of the frequency you are going to simulate IBIS model at.
>
> Hope this helps.
>
> Best,
>
> Tim
>
>
> -----Original Message-----
> From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx]
> On
> Behalf Of Vishwanath Elechithaya
> Sent: Friday, May 28, 2010 11:24 AM
> To: si-list@xxxxxxxxxxxxx
> Subject: [SI-LIST] Re: Looking for an algorithm to clip waveforms
>
> Hi Kumar,
> Tuning C_comp is again a issue that I need to do a iterative process. The
> primary idea is to make a good algorithm for time-shifting, so that the
> algorithm be implemented under a tool. I have clipped the timing manually
> and was able to come to a point with all the waveforms matching to spice.
> however, I am not able to fit everything into a logical equation to be
> implemented in a tool.
>
>
> On Fri, May 28, 2010 at 7:13 PM, C. Kumar <kumarchi@xxxxxxxxx> wrote:
>
> > if it is possible you shuld et the oriiginal silicon model and develop a
> > simple behaviour model, use the ibis model with the vt tables removed and
> > tune the c_comp values to match the output response
> >
> >
> > ------------------------------
> > *From:* Vishwanath Elechithaya <elechi@xxxxxxxxx>
> > *To:* si-list@xxxxxxxxxxxxx
> > *Sent:* Fri, May 28, 2010 8:53:46 AM
> > *Subject:* [SI-LIST] Re: Looking for an algorithm to clip waveforms
> >
> > Hi Kumar,
> > This is not a good solution as the waveforms no more match! if I remove
> > the VT tables.
> > Regards,
> > elechi
> >
> > On Fri, May 28, 2010 at 3:54 PM, C. Kumar <kumarchi@xxxxxxxxx> wrote:
> >
> > > viswanath:
> > > you will save yourself lot of time by just removing the rising and
> > falling
> > > wave form tables. These tables were used to represent staged switching
> in
> > > old cmos drivers.
> > > At DDR3 speeds they are mainly legacy.
> > >
> > > ------------------------------
> > > *From:* Vishwanath Elechithaya <elechi@xxxxxxxxx>
> > > *To:* si-list@xxxxxxxxxxxxx
> > > *Sent:* Fri, May 28, 2010 4:47:27 AM
> > > *Subject:* [SI-LIST] Looking for an algorithm to clip waveforms
> > >
> > > Hi All,
> > > while generating models for a fast DDR3 Cell, I came across a problem
> of
> > > over-clocked cell problem of the IBIS models not able to cope up the
> > > frequency requirements. I came across some topics here about the time
> > > shifting of the waveforms. I have tried out by manually clipping the
> > > waveforms so as to match the spice w.r.t IBIS, with a bit of success.
> > > however, the values for the clipping I am not able to put them in a
> > > generalized algorithm for various over-clocked models. I hope that many
> > of
> > > you may have come accross this situation, and would like to know if any
> > > algorithm exists or innovated to actually find the amount of clipping
> > that
> > > needs to be applied to the models.
> > > Regards,
> > > elechi
> > >
> > >
> > > ------------------------------------------------------------------
> > > To unsubscribe from si-list:
> > > si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
> > >
> > > or to administer your membership from a web page, go to:
> > > //www.freelists.org/webpage/si-list
> > >
> > > For help:
> > > si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
> > >
> > >
> > > List technical documents are available at:
> > > http://www.si-list.net
> > >
> > > List archives are viewable at:
> > > //www.freelists.org/archives/si-list
> > >
> > > Old (prior to June 6, 2001) list archives are viewable at:
> > > http://www.qsl.net/wb6tpu
> > >
> > >
> > >
> > >
> >
> >
> > ------------------------------------------------------------------
> > To unsubscribe from si-list:
> > si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
> >
> > or to administer your membership from a web page, go to:
> > //www.freelists.org/webpage/si-list
> >
> > For help:
> > si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
> >
> >
> > List technical documents are available at:
> > http://www.si-list.net
> >
> > List archives are viewable at:
> > //www.freelists.org/archives/si-list
> >
> > Old (prior to June 6, 2001) list archives are viewable at:
> > http://www.qsl.net/wb6tpu
> >
> >
> >
> >
>
>
> ------------------------------------------------------------------
> To unsubscribe from si-list:
> si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
>
> or to administer your membership from a web page, go to:
> //www.freelists.org/webpage/si-list
>
> For help:
> si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
>
>
> List technical documents are available at:
> http://www.si-list.net
>
> List archives are viewable at:
> //www.freelists.org/archives/si-list
>
> Old (prior to June 6, 2001) list archives are viewable at:
> http://www.qsl.net/wb6tpu
>
>
>
>
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List technical documents are available at:
http://www.si-list.net
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
Other related posts:
