SOS! SAVE ME! cdsSpice could not extract netlist: ERROR: NODE/NET IS NOT ON KEEP LIST AND NOT IN WSF

Dear all,

This time I am using IC5.0 from cadence. I want to extract netlist to feed
into nanosim for post-layout verification...

The design is about 5000 cells(statistics from Silicion Ensemble). I
successfully obtained extracted view in icfb.

Then I invoked Analog Artist Environment and change from the default
"spectre" format to "hspiceS" format. And then "create final netlist",

but after waiting for one day and one night, I got a bunch of :

\o ERROR: NODE/NET IS NOT ON KEEP LIST AND NOT IN WSF FILE
\o C749063 8284 14963 106.280929661141e-18 M=1.0
\o ^
\o ERROR: NODE/NET IS NOT ON KEEP LIST AND NOT IN WSF FILE
\o C749067 4347 15763 787.456302574018e-18 M=1.0
\o ^
\o ERROR: NODE/NET IS NOT ON KEEP LIST AND NOT IN WSF FILE
\o C749069 4347 8284 408.707690731185e-18 M=1.0
\o ^

The obtained netlist "hspiceFinal" is only partial, no transistors were
listed...

To be frankly, in the past 20 days, I never could get through this last
step- netlist extraction.

Anybody familiar with these errors? Can you please kindly save me?

Thanks a lot,

-Walala

 

Hi Walala,

I found a couple of instances of this error message in reported service
requests; both were when very large parasitic cellViews were being netlisted
into a socket simulator.

Unfortunately the various tricks used to try to solve this only seemed to
solve the memory or timeout problems that might sometimes occur - but
this problem seemed to remain.

In both cases the customers failed to respond after lengthy attempts to
get to the bottom of the problem; either meaning that they'd solved
the problem somehow or given up...

One positive thing is that this error comes from cdsSpice and hence would
go away with a direct netlister (such as spectre, or the upcoming hspiceD
interface).

Other than that, since I've not seen this before, it would appear that something
has got confused in cdsSpice. I think it would be hard to get to the
bottom of without seeing it - so I can only suggest that you follow up
via whatever support channel you have.

Is it possible to netlist in "spectre", and then use a perl script or similar
to convert into a format that is suitable for nanosim?

Regards,

Andrew.

 

Dear Andrew,

Thanks for your answer. And I hope you are having a great weekend.
(for me the weekend is as sore as I previously had, trying to find a way
through out of a virtually exponential search space via manual trial and
error week and week for the same purpose: get the netlist and run nanosim to
get power...)

I guess the laest service request you found was turned in on Friday: our
university CAD manager filed it finally. So you are saying that our request
is almost doomed, since previous similar cases were not successful? :=)

Thank you for pointing out the "spectre" way. I tried and "spectre" worked.
It can successfully extract the netlist! But now the service request turned
to Synopsys. After several rounds, their service person admitted that
nanosim is too picky to take in the "spectre" format netlist... They have no
solutions.

Ooops, now I am opted to use the last resort: your advice to do a conversion
by perl. Before trying out that, I have several questions:

a. In icfb Analog Artist, do I choose "simulation host" to be "spectre" or
"spectreS"? What 's the difference?
b. Is there any existing tool to do the conversion from "spectre"/"spectreS"
to "hspiceS"? I heard about a tool named "SPP", but I am not sure if it will
work for our purpose?

Thank you very much and I really appreciate your help. I also appreciate
that Cadence has this active newsgroup to give a communication channel aside
from the normal "many students" <==>"one unlucky CAD manager" <==> "unlucky
Cadence support personel facing many CAD managers" channel.

-Walala

 

Hi Walala,

Some responses below...

Actually, I hadn't seen that - the search I was using is only indexed overnight,
and so it wouldn't have shown up at the time I did the search. That said, I
only did a quick search and possibly with input of R&D I could have got
further. I would say that you stand more chance of getting this resolved
if you have a reproducable testcase which you can share with Cadence - there's
nothing like being able to reproduce the problem to pinpoint the cause!

spectreS is the _old_ interface to spectre. It is a "socket" netlister - which
means that it netlists in cdsSpice's syntax, and uses cdsSpice to generate
the final netlist. This is done in order to take advantage of cdsSpice's macro
language - which allowed parameterisation of the netlists before spectre
fully supported this.

It's really obsolete now - spectre produces a direct netlist and misses out
cdsSpice. Quite likely, spectreS would suffer from the same problems
you've seen with hspiceS.

'fraid not. spp goes the other way - SPICE syntax into spectre syntax.
One of my colleagues has a utility to go from spectre to Verilog-AMS,
but that won't help you either...

No problem - glad to be of assistance (where I can).

Regards,

Andrew.

 

Dear Adrew,

Thank you very much for your suggestions! After taking your advice and after
looking at the two netlist formats, I think writing perl script to make a
conversion might be a good way... since the instances in two netlists have
one-to-one correspondence, which makes things a lot easier...

Have a great weekend, and I will report further progress once I get some
results,

-Walala