How to run monte carlo using spectreS

Dear Cadence,

In analog artist, must I use spectre (Direct) instead of
spectreS in order to run monte-carlo simulation?

There is no "Model Libraries" menu item in spectreS mode
so how can I specify the file where I put the statistics{}
block?

There does exist a menu item "monte carlo" in spectreS mode, so
it doesn't make any sense if there is no way to use it!

Thanks,

Cao

 

The spectreS mode (if my memory is correct) uses the old way of doing monte
carlo, rather than the built-in way of using the montecarlo analysis within
spectre.

Before IC443 (which was released ~6 years ago), spectre did not have a native
monte-carlo analysis - many simulators were in a similar position. So spectreS
uses cdsSpice to generate all the random numbers and generate an evaluated
netlist for each point run (which as you can imagine makes it less than optimal
in terms of speed!). The syntax for specifying the distributions is using
cdsSpice syntax, and does not therefore use the statistics block.

Since IC443, spectre has a native montecarlo, and uses the statistics block. At
the same time, a direct interface was added "spectre" to replace "spectreS", and
that is really the right way to use spectre.

As mentioned many times on this news group, spectreS is obsolete. In fact it's
dead and buried - IC5141 is the last release that it is present in.

If you're going to use monte carlo, I would strongly recommend using it in the
"spectre" interface.

Regards,

Andrew.

 

Hi Andrew,

Many thanks that you always answer questions on this newsgroup.

There are several reasons that I try to avoid switching to the
spectre direct interface, hopefully you can help me find out
ways to solve them:

1. Everytime I change the schematic, I must "Check and Save" or the
simulator will not run. Why can't it automatically do "check&save"?

2. I am using the NCSU Cadence Design Kit
http://www.cadence.ncsu.edu/CDK.html
It seems that the schematic libraries included in this kit only
supports spectreS. If I use other libraries such as analogLib
to be able to use spectre direct,
a. I need to change every component in my existing schematic
from the library NCSU_Analog_Parts to equivalent cell in
analogLib which is a lot of work
b. If I use analogLib, the property page for MOSFET cannot
automatically calculate W and L from grid units and diffusion
area from W and L.

3. The output window that pops up when I run spectre direct is bothersome.
Why can't it display output in the icfb window?

4. Besides, I don't feel spectre is much faster than spectreS. I even
think it is slower because while spectreS is running in separate process
and it does not freeze the icfb GUI in anyway, during start-up of
spectre direct simulation the icfb GUI will freeze for several seconds.


Thanks,

Cao

 

It might seem trivial to do this, but in a big design, it can take longer to
figure out what needs checking and saving (down through the hierarchy) than
it does to actually perform the netlisting. So this was part of streamlining the
simulation.

And checking-and-saving all the time isn't an option either, since that then
unnecessarily re-netlists when nothing changed.

You don't actually need to save - but you need to check. It doesn't take
much to get into the habit of hitting x (if that's your bindkey for check)
after making a change before simulation. After all, you knew you just changed
something!

Well, that means it must be quite old, given the fact that spectreS has
really be obsolete for 6 years. I can understand your frustration.

The CIW was never designed for mass display of information. In the past
we used to have issues with a non-converging simulation taking several
minutes to tell you why, when the simulation failed in seconds. Displaying in
a separate window is much more efficient.

If you don't like seeing the output pane, you can always disable it.
Setup->Environment - turn off automatic output log (there's a cdsenv)

You must be in a minority in that view. It does not normally freeze DFII - it
does take a little time to netlist (far less than spectreS). It doesn't then
have recreate a whole new netlist every time a parameter changes, and also
spectre can remain running - so things like parametric analysis and monte carlo
are much, much, much faster.

I can't really think of anything that is slower than spectreS with spectre.

Best Regards,

Andrew.