I'm about to simulate a power amplifier with spectre using mmsim 10.1 and cadence ic 5141. I want to analyze if it oscillates in transient simulations. Therefore I gave an inductor a initial condition which is not 0 so that there is a stimulation.
I read the following recommandation to simulate oscillators in a cadence blog post: (/blogs/rf/archive/2012/12/20/mmsim12-1-speed-up-transient-analysis-of-crystal-oscillators.aspx)
"For Integration Method, select traponly. By default, when errpreset=conservative, the integration method is gear2only. However, gear2only can dampen oscillations and shouldn’t be used – you want your oscillator to oscillate! So, I recommend using either trap or traponly when simulating crystal oscillators."
So I choose traponly and got problems with conversion. (The timestep was reduced to a few attoseconds) Even if I choose liberal as error preset. The simulation result is shown here:
First I thought there is an oscillation and zoomed in, which you see here:
Then I thought that there are some issues with undersampling or something like that, but if you look at the timestep, then the frequency would be higher than 500 GHz because the period of this triangle signal is about 2ps.
But I'm using a SiGe Technology with a transit frequency of 200 GHz.
Then I simulated the circuit using gear2 method and got the following result:
Which seems to be ok. Because nothin is oscillating at the end and the oscillation from my inductors initial condition is dampened very fast.
It also works if I set the errorpreset to conservative as I do usually.
I also read in the spectre reference about the different envelope integration methods like gear2, trap, etc. But I only saw that there are different presets in the timesteps, tolerances, steadyratios and so on which doesn't help me in understanding when do I have to use which method? And how do they work. What is the problem with using traponly?
I hope anyone can help.
What you are seeing here is a phenomenon called trapezoidal ringing. It's a numerical oscillation - the indication that it is numerical is that the oscillation is between adjacent timepoints (if you show all timepoints in the waveform, there should be none between the peaks of the oscillation).
The typical ways to cure trapezoidal ringing are to use a gear method (e.g. gear2only) because this introduces (amongst other things) a small amount of numerical damping, or you could tighten reltol. The downside of gear2only is that the small numerical damping can artificially damp out oscillation, and hence it can make it hard to simulate oscillators and also it can make unstable circuits appear more stable.
So you might also try tightening reltol (e.g. to 1e-5) and see if that helps.
However, maybe gear2only is fine with your circuit - it rather depends on what you are expecting!
You can find some more background information in http://www.kenkundert.com/docs/bctm98-MSsim.pdf (starting on page 18, with trapezoidal ringing explained on page 22) and in Ken Kundert's book "The Designer's Guide to SPICE and Spectre", see http://www.designers-guide.org/Books/dg-spice/index.html.
I would also like to point out that the Spectre version you are using is rather old (the current version is MMSIM 13.1).