The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> A problem with pss analysis
https://designers-guide.org/forum/YaBB.pl?num=1143183627

Message started by goldendawn on Mar 23^rd, 2006, 11:00pm

Title: A problem with pss analysis
Post by goldendawn on Mar 23^rd, 2006, 11:00pm

I am doing postlayout simulation with a Mixer ,using the tool of Cadence spectre. The parasitic parameters including resisters and capacitors, have been extracted with Calibre. When pss analysis is executed, the memory is rapidly run out, and the simulation is interrupted with fatal error. But the tran analysis can be executed normally. the scale of the circuit is small, and the quatity of memory is 4G, so i think there is something wrong with my setting of simulation but not with the lack of memory.
thanks a lot.

Title: Re: A problem with pss analysis
Post by Andrew Beckett on Mar 30^th, 2006, 5:18am

Without giving any clues as to what you've done, it's hard to tell. The new front end (in MMSIM60 by default, or on in IC5141 by providing the +csfe
switch) tends to reduce the memory the parser takes - but it may be down to the PSS. You don't give any clue as to the frequencies you have, the PSS settings, the size of the circuit, etc, etc.

With no information to go on, we can't help you much!

Andrew.

Title: Re: A problem with pss analysis
Post by schehrazi on May 25^th, 2006, 5:48pm

Andrew is right, however, if you try to selectively save your signal at some specific nodes instead of saving all signals, that might help.

Title: Re: A problem with pss analysis
Post by Geoffrey_Coram on May 26^th, 2006, 5:34am

Doing a PSS analysis using shooting-Newton requires the simulator to store the circuit matrices for every point along the period, which it then uses to compute an updated initial condition. Even if you only ask it to save certain nodes in the converged solution, the memory will still be used while computing the solution.

Another way to compute the PSS solution is to simply run a period and check if initial=final, and if not, run another period and try again. In this case, the simulator only has to store the V-vector for the initial condition (in addition to all the things it usually stores while computing a regular transient analysis). The only difference between this and transient is that the simulator detects convergence to a periodic solution, rather than the user having to run the analysis and manually check for convergence.

Andrew, do you know if Spectre can do this sort of PSS, and if the memory requirements would, in fact, be reduced?

Title: Re: A problem with pss analysis
Post by Andrew Beckett on Jul 20^th, 2006, 1:42pm

Geoffrey,

Apologies for the lateness of my reply - I've been tied up with so many other things recently, I've had no chance to monitor the Designer's Guide.

No, SpectreRF does not provide a means of doing what you're suggesting. I'm not sure it would be useful, given the fact that the matrices over time are needed for the small-signal analyses.

If the circuit is weakly non-linear, and especially if it is a multi-tone simulation, using flexible balance can save a lot of memory, since you only need to store the harmonic coefficients for a small number of harmonics for each voltage and branch current (put simply).

Also, using QPSS can help if it's a multi-tone simulation, because even with shooting, the solution is partly in the frequency domain, and the time domain information is only needed for the "large" signal.

Andrew.