Cadence IC6/5 automated simulation using Spectre

[guitargore]

Hello,

I am trying to do genetic algorithms research using Matlab and Cadence IC6/Spectre. I will be generating a matrix in Matlab that defines the connections in a circuit. I want to somehow pass this information into Cadence to define the circuit connections, simulate the circuit performance, and return the results.

It is important to remove as much human involvement as possible.

I have no idea where to begin working on this. Any input would be greatly appreciated.

 

[dick_freebird]

Seems to me you'd be well off to forget Cadence and just use any available
SPICE, preferably one with a decent scripting language (Spectre is about the
weakest and most useless in this respect, you have to use Ocean which is
verbose and syntactically fussy). It's not like this matrix of yours is going to
be of any use to Cadence tools as a design input anyway. If you're using a
standards based model (like BSIM) then you might pull the model out of the
design system, get it into a regular SPICE and go to town. You ought to be
able to concatenate a standard header, your gen'd up body and standard
footer and punt a batch mode run and pick the result off the output file
with Matlab, PERL or some combo.

Unless you need the fancy RF analyses or something. Or unless your professor
is Cadence's love slave.

 

[guitargore]

That is probably the best place to start this project. The reason we want to eventually move to Cadence and Spectre is because we do our IC design in those tools. That part is more of a long term goal rather than an immediate requirement.

I believe by " concatenate a standard header, your gen'd up body and standard
footer" you mean to generate the .CKT file for spice. For a newbie, could you elaborate on the batch run or is that something I might find in the documentation of the spice simulator (maybe NG Spice)?

 

[dick_freebird]

It would be something like

ngspice <myCatFile.in >&myOutputFile &

kicked off in UNIX / Linux under control of whatever your master script
is; you'd have the script monitor the job queuing I suppose, and maybe
make uniquely named input and matching output filenames for an orderly
data-pile (just on the chance that some runs bomb, you can repeat
only the bad ones rather than having to run the whole deck over and
over. Just brought back a whole bunch of memories of the good old,
bad old days. t*v*mccnt looped runs and grind your own statistics.