How to get smooth and accurate output waveform without increasing simulation time

A waveform does not need to be smooth to be accurate. It's the points that matter, not the lines between them (since the simulator solves at the points). You could try tightening the reltol to 1e-4 (or set errpreset=conservative, although that will probably slow the simulation down more than just tightening reltol). More points are likely to take more time though.

Andrew.

In addition to Andrew’s suggestion if you prefer not to change “errpreset” nor reltol, if you include a limit for maxstep you will get a voltage at least every maxstep interval. There is a default value for maxstep defined by the length of your simulation and the value of errpreset you select. However, you can define maxstep to be a lower value. As Andrew also noted, although lowering the value of maxstep may not increase the simulation time as much as a decrease in reltol or errpreset, it will increase simulation time.

Shawn

Thanks Shawn, although I should also point out that tightening maxstep is a poor way of improving accuracy - it tends to slow down the simulation more than tightening reltol for a similar level of accuracy, because you are forcing tight time steps all the time even when they're not needed.

Andrew. 

Dear Andrew, Shawn,

Thanks a lot for your suggestions.

I am curious to know the difference between tightening reltol and the errpreset. Does errpreset tighten other parameters apart from relotol.

Kind Regards,

If you type "spectre -h tran" you'll see in the text further down that it gives a table of what errpreset changes. One key difference is that it will change relref to alllocal (it also loosens lteratio a little which will prevent quite so many time steps from being taken too).

Andrew.

Dear Andrew,

"One key difference is that it will change relref to alllocal (it also loosens lteratio a little which will prevent quite so many time steps from being taken too)."

Does the above mean tightening errpreset takes less simulation time than tightening reltol ?

Probably not, because changing relref from sigglobal to alllocal is likely to increase simulation time as this will improve accuracy. It's very hard to be definitive about these things as the complexity does tend to be circuit dependent.

In general I'd normally suggest going from moderate to conservative before tightening reltol, but sometimes you don't necessarily need a local reference. Predicting speed of simulation is a complex art... (certainly too much to explain quickly in a forum; you might want to read Ken Kundert's book The Designer's Guide to SPICE and Spectre (the book's a bit old now, but it still remains highly relevant and useful).

Andrew.

Thanks a lot Andrew.

Thanks a lot Andrew.

Thanks a lot Andrew.