Hello All,

I am working on an LC oscillator design and trying to simulate the phase noise using SpectreRF HB/HBNOISE analysis.  This has worked well for me in the past, but now with a newer version of MMSIM I see these messages in my output log during the HBNOISE analysis:

Warning: GMRES did not converge to the desired tolerance of 0.0001. It actually converged to 0.000129598. The solution may not be accurate but it's saved anyway.
Warning: GMRES did not converge to the desired tolerance of 0.0001. It actually converged to 0.00288411. The solution may not be accurate but it's saved anyway.
   hbnoise: freq = 10 kHz         (0 %), step = 10 kHz          (0 %)
Warning: GMRES did not converge to the desired tolerance of 0.0001. It actually converged to 0.000131274. The solution may not be accurate but it's saved anyway.
Warning: GMRES did not converge to the desired tolerance of 0.0001. It actually converged to 0.00405571. The solution may not be accurate but it's saved anyway.
   hbnoise: freq = 100 kHz     (33.3 %), step = 90 kHz       (33.3 %)
Warning: GMRES did not converge to the desired tolerance of 0.0001. It actually converged to 0.000128369. The solution may not be accurate but it's saved anyway.
Warning: GMRES did not converge to the desired tolerance of 0.0001. It actually converged to 0.00204469. The solution may not be accurate but it's saved anyway.
   hbnoise: freq = 1 MHz       (66.7 %), step = 900 kHz      (33.3 %)
Warning: GMRES did not converge to the desired tolerance of 0.0001. It actually converged to 0.00219455. The solution may not be accurate but it's saved anyway.
   hbnoise: freq = 10 MHz       (100 %), step = 9 MHz        (33.3 %)

Does anyone have experience with these warnings and how to improve GMRES convergence?  Also, I am running MMSIM13.1.

Thanks,
Tim

Hello Andrew,

Thank you for your response.  We are not running MMSIM13.1 hotfix.  We will install that and see if this solves the problem.  If not, we will contact Cadence support.

Thanks again,
Tim

Hi Tim,

You might want to try this parameter on hbnoise:

Code:

44	krylov_size=200   The minimum iteration count of the linear matrix solver used
				  in periodic small-signal analysis. After reaching
				  krylov_size  iterations, the iteration is forced to
				  terminate because of the poor rate of convergence. Increase
				  the krylov_size if the simulation reports insufficient norm
				  reduction errors in GMRES. 

You'll need to set this via the additionalParams field on the hbnoise options form. Not sure about whether you can trust the results as is (it's reasonably rare to get this GMRES message) - and that's why it is important to work with customer support on this.

A little bit of info on GMRES can be found at http://en.wikipedia.org/wiki/Gmres

Regards,

Andrew.

A better forum for questions like this is http://community.cadence.com/cadence_technology_forums/f/33. I did a quick search on http://support.cadence.com/ and did not find any other information about the krylov_size parameter than what was given by Andrew. Solution 20184596 (http://support.cadence.com/wps/mypoc/cos?uri=deeplinkmin:ViewSolution;solutionNu...) suggests that the convergence problems are fixed in MMSIM 13.1 ISR5.