Spectre vs Nanosim

Dear all,

I have a SRAM circuit (/circuits with different wordlengths from 8
to 48) which I need to simulate and do the power analysis. Initially I
thought I could do the power analysis by simulating the circuit with
Spectre. But my supervisor said that it could take many days to
simulate one such circuit and I have to do these simulations on more
than one circuit. So he suggested me to explore the possibilitys with
nanosim. Is it fesible to do such a power analysis on spectre or else
would you suggest me to use nanosim instead. How much time will it take
to simulate a SRAM circuit with wordlenghts 8, 12, 16, 32 and 48. Your
inputs will be very much apprecieted.

Thanks and Regards,
Kamesh.

 

I forgot to add one thing.The SRAM is a layout level circuit in the
above.

Thanks once again for your time,
Kamesh.

 

I don't know how much time you will save, but you will save time. Depending
on the accuracy setting of nanosim you will save a lot of time. But beware:
You trade time vs. accuracy. That accuracy may sometimes be so bad that you
may be fooled into wrong conclusions.

Nanosim and its GUI offer some nice statistical functions to find the power
consuming devices. But again beware: Never trust a statistic you haven't
forged yourself. You have to take some time to get familiar with those
functions or you may again make the wrong conclusions.

I finally backed all nanosim runs (that took 1h each) with one final spectre
simulation (that took 3d) to verify the nanosim runs.

Your Milage Will Vary

 

Hi Kamesh,

I have not used nanosim, but have used other fast spice simulators
such as powermill, starsim etc.
Actually if you would like to get a good accuracy and performance with
spectre,
you could actually parametrize your SRAM and create a critical path
netlist(You do
need to know the metal capacitance, architecture of your
decoder/wordline drivers
and the control logic, apart from your memory cell core).
If you are going to use nanosim, you could also sample some of your
nanosim
simulations and compare with spectre results for crosscheck on
accuracy.
In which case you might also have to identify the blocks in your
SRAM(such
as sense amps/precharge circuit etc) and use special facilities in
nanosim to
treat them as pure analog circuit with real spice simulation.
While trying to run the full simulation with spectre, it might help to
use the initial
condition file from nanosim, to avoid any convergence issues.

I hope you find these suggestions helpful.

Regards
Raman

 

AVOCAD simulates (http://www.avocad.com/) simulates logic circuits on
transistor-level very well

 

AVOCAD simulates (http://www.avocad.com/) simulates logic circuits on
transistor-level very well

 

Thank you very much for your valuable insights Svenn and Raman. Based
on your suggestions the following are the alternatives that I can see
and I am summarising them in here.

1) I can use Nanosim only after making a comparison of the power
consumed by SRAM (say of wordlength 8) with that of the result
obtained from Spectre simulation for the same. This is done just to be
sure that the Nanosim results are within the acceptable range.

2) Geoffrey_Coram of designers guide has suggested that it may be
possible to analyze only one bit row with a multiplicity factor. He
says that this could make a dramatic difference in the speed of the
simulation. [Thanks Geoffrey]

I am thinking about the above two alternatives and will discuss them
with my supervisor and start working on it soon.

Thanks and Regards,
Kamesh.