I try spectreX to speed up the simulation.
With "preset" = MX/AX, the tran waveform displays trapezoidal ringing.
With "preset" = CX, the ringing disappear, but paying more time and license.
Is there any way to avoid trapezoidal ringing with MX/AX in spectreX? (I suppose they are already analog precise enough)
The tran algorithm is set gear2only.
When i use APS and errpreset = moderate, there is no ringing.
sjwprcker said:Is there any way to avoid trapezoidal ringing with MX/AX in spectreX? (I suppose they are already analog precise enough)
Did you happen to see the note entitled "Getting the Most Out of Spectre X" at the Cadence On-line support site URL:
Page 15 and 16 detail how to use the preset override feature with Spectre X. I put together an example of how to set the relevant GUI and the subsequent relevant sections of the netlist and spectre.out file showing the use of Spectre X with the preset override feature set to "method" and the method set to "gear2only" in the hope it might help. I've placed the note at URL:
I hope this helps sjwprcker!
Thanks for your feedback, it works.
Can you give more info why gear2only is proposed here, and why not other integration methods? Before SpectreX, we generally use trapgear2 to avoid trapezoidal ringing..., any rational for gear2only?
I was happy to red ti worked for you - great!
sjwprcker said:Can you give more info why gear2only is proposed here, and why not other integration methods?
I cannot speak for Cadence's proposal specifically with respect to Spectre X, but the second order gear integration method (also called Backward Differentiation formulae or BDF, and actually proposed by Dr. C. W. Gear in 1971) is known for its stability (i.e., immunity from the numerical ringing seen in trapezoidal algorithms). Hence, in many transient simulations, it is my choice among integration algorithms.
sjwprcker said:Before SpectreX, we generally use trapgear2 to avoid trapezoidal ringing..., any rational for gear2only?
As shown in the Spectre reference manual, the "trapgear2" choice for an integration algorithm does not preclude the use of a trapezoidal algorithm, but allows the simulator to choose between Backward-euler, Trapezoidal, and Second-Order Gear as shown at the On-line support URL:
and included below for your convenience. Hence, its choice does not enforce the use of a second-oder gear algorithm exclusively.
Does this help at all sjwprcker?
There is some additional background information at https://support.cadence.com/apex/techpubDocViewerPage?path=spectreRFTheory/spectreRFTheory20.1/chap3.html#pgfId-1015935 . One point to be aware of is the fact that Gear's backward difference method exhibits artificial numerical damping.
Your answer is quite clear, thanks a lot!
Thanks for your comments, they are also useful information.