High Q XOSC - PSS shooting engine: erroneous results

Unknown said:

I am designing a simple high Q 16M XTAL + INV oscillator in MMSIM 14.10.576 (I think this is the latest version?) using PSS with the shooting engine. The reason I went for the shooting engine rather than the hb is that there is a significant amnount of distortion in the waveforms. On top of that I found that the hb engine gives convergence problems in some of my corners (typically 2 or 3 of my 22 corners will fall over). These errors can be fixed by tightening reltol and vabstol as recommended in the SpectreRF documentation, however, doing so will make other corners not converge!

Therefore I switched to the shooting engine which I was hoping would give me reliable results over all coners. Unluckily, I find even though the simulation always converge, I get strange results 1 corner (very low amplitude steady state). Changing simulation engine parameters will not help, the results will always be very similar. I checked the behaviour of the design in that particular corner with a transient noise simulation and the oscillator reaches the expected steady state with a normal amplitude. What it is more when I use the hb engine in that corner the results are the right one :s

So I guess my questions are:

- Any insight in why the shooting engine is giving the wrong steady state solution? What can I tweak to improve solution accuracy?

- Is there any way to rerun a set of corners in which several BUT not all the analysis in a test have failed AND add the results together to the corners Job? I know it is possible to rerun erroneous corners but in order to be able to do so all the analysis should have failed. So far I am putting a note to the corners Job but it is messy

 

=> It would be helpful if you posted your shooting pss and hb analysis statements as well as your options statements.

Not having that information, I'll offer some generic suggestions...

In hb analysis, there are autotstab and autoharmonics features which are very helpful in cases such as this.   I would try that first with errpreset=moderate or conservative.   Don't tweak any other options - leave at defaults.  You will need to set oversample.  Depending on the level of distortion in the signal, try oversample = 4 or 8.

In shooting, use the new autotstab feature.   If errpreset=conservative doesn't work in shooting (too many convergence issues) and errpreset=moderate isn't accurate enough, try tightening:   reltol=1e-5, vabstol=3e-8 errpreset=moderate

Best regards,

Tawna

Hi again,

Many thanks for your reply. My analysis statements are:

=> pss

analysis('pss ?fund "16M"  ?harms "100"  ?saveinit "yes"  ?oscana t  ?p "/xoscp"  ?n "/gnd!"  ?oscic "lin"  ?method "traponly"  ?ppv ""  )

=> hb

analysis('hb ?saveinit "yes"  ?autoharms "yes"  ?autotstab "yes" ?oversample list("1")  ?fundfreqs list("(16M)")  ?maxharms list("5")  ?errpreset "conservative" ?oscana t  ?p "/xoscp"  ?n ""  ?oscic "lin" ?oscmethod "onetier"  ?pinnode ""  ?ppv ""  ?tstabmethod "traponly"  )

For the hb analysis I was using autotstab and autoharmonics, I ll try to increment the oversampling factor to 4 or 5 as suggested.

In pss I tried with errpreset=conservative reltol=1e-4, vabstol=1e-6 first and then with errpreset=conservative reltol=1e-5, vabstol=3e-8. In both cases I ended up running into convergence issues in some of the corners (different corners for each configuration). I will try using errpreset=moderate

Best,

Alberto