仿真遇到trapezoidal ringing，输出电压越来越高，都到几十V了，怎么解？

hehuachangkai

有什么设置可以消除这个问题？ 精度、算法都试了，没有办法。目前大电阻小电容还没有试

黑化的玛奇朵

老哥，解决了吗，求求解决方案，谢谢啦

zwzdqhaust

cadence的手册里写的：在瞬态分析中， 梯形规则响铃很烦人，而在PSS分析中，它会导致射击迭代在收敛之前停止。您可以通过将方法从traponly更改为trap、gear2或gear2only来解决这个问题。

cx990324

下面有个我在其他地方查到的，供参考

cx990324

Gear method is better than others to get higher accuracy, also in Spectre.
We already know the above in Hspice but it was not clear, at least to me, in Spectre, because it just says ‘default’, but what is the default?, what is his ‘conservative’?
Now I understand that when we scpecify ‘conservative’, then his ‘default’ will be set to ‘gear2only’. Here, also in Spectre, gear is better than others, even though the help file says “trapezoidal is used when we need high accuracy” and “gear can make system more stable than they really are.”

If you used trap or traponly, try and select gear2, gear2only or even euler.
For the differences, study e.g. Kenneth S. Kundert’s book “The Designers Guide to SPICE & SPECTRE”, especially the 4.2.2 chapter on “Characteristics of the Integration Methods” (about 25 pages).
Here just a few clippings: “trapezoidal rule exhibits strong point-to-point ringing.” (p.135); “backward Euler exhibits heavy damping” (p.147); “trapezoidal ringing” (p.150); “To avoid trapezoidal ringing, use Gear2” (p.155).
If you set Spectre’s errpreset property to conservative, the gear2only method is used by default. I’d suggest to use this one first