Stephen,

 Try running a periodic stability analysis and look at your gain and
phase margin. It seems to work for the this type of design. The other
thing to explore is how complete is your model does it include interconnect
parasitics, packaging parasitics, and board parasitics? Are you using an
ideal capacitor model or a does the model include ESR and ESL? ...?

                                                       Best Regards,

                                                         Sheldon

Stephen,

  Yes, periodic stability analysis is part of SpectreRF. Unfortunately, the
Charge Pump is driven by a periodic signal and has a periodic response.
So it is well suited to be analyzed by SpectreRF's time-domain PSS analysis.
I do not believe that there is an alternative solution unless you can get
the state-average model simulations to correlate with measured results.
In that case, you could use Spectre's stability analysis to analyze the
loop's characteristics. BTW, when you analyzed the loop stability using
the model in the article by Thiele and Bayer  did you include the capacitor
parasitics in the stability analsysis?

                                                                Best Regards,

                                                                   Sheldon

Stephen,
If transient analysis isn't showing instability (i.e. ringing during start-up or a step change in load current) I wouldn't think PSS would either.  Since you are getting similar results with more series ESR, perhaps it is the bond wire inductance that is bumping up your feedback signal - especially if you didn't kelvin connect the feedback signal to the Load capacitor.  DC current loading may play a role, and it wouldn't surprise me if some large disturbance wouldn't set up a non-linear oscillation.  Other than that, with all those inversions and series resistances, I can see where unaccounted parasitics could generate secondary poles.  My first guess would be the parasitic cap at the inverting node of the amplifier, then across R1.  The loop M7/M6 looks odd to me too.  

The diodes probably have a parasitic bipolar associated with them.  It would get real exciting if your diode actually doubled the current somehow.

So.... I'm guessing you need to figure out what parasitic you are missing instead of PSS.

Sometimes you can force the output to a known voltage with a power supply and monitor current for oscillations.  If you can find regions (i.e. output levels corresonding to different load currents) where it is stable, this might help point you to the right answer.

I briefly looked at the paper you referenced.  I expect you know this, but that is a clever scheme that operates differently than your normal charge pump.  They use the "Flying capacitor" as a level shifter, not a "Flying Capacitor."  If you are using the flying capacitor in the traditional way to actually transfer the charge then the stabilizing parameters are completely different.  Still, I'd think simulations would predict what you are seeing.  

Hope this helps at least a little,
Rob

That's great that the FIB worked.  It is hard to guess what parasitic was responsible without having more detail about the circuit and what you fibbed out.

I've only skimmed the paper so I may have missed something, but my "traditional way" would correspond to their "voltage mode."  In the "traditional" mode you are using the flying capacitor to transfer the charge.  for example, during "charge phase" you put on a controlled amount of charge, and during the "discharge phase" you discharge the flying capacitor by putting it in parallel with the load capacitor.  When you do that, the size of the capacitor as well as the charging time figure into the stability criteria.  The implication of Fig. 2 is that using a current source changes the stability criteria... this isn't necessarily true because you can use a current source to charge the capacitor during the "charge phase."  

I like the idea presented in the paper... but at first glance it looks more like a linear regulator with a series capacitor used for level shifting so that the feedback amplifier can operate beyond the rails.  Therefore, stability analysis should be similar to that of an LDO.

I hope you'll let us know what it was that was causing the instability.