Re: [SI-LIST] : Question about Gear vs Trapezoidal methods for simulations within HSPICe

J. Eric Bracken (bracken@ansoft.com)
Thu, 05 Mar 1998 20:28:52 -0500

Hi,

Trapezoidal integration is the workhorse of Spice and Spice-like
simulators. Unfortunately, when you approach the steady-state value
of a waveform, it has a tendency to oscillate. I believe that Spice2
switches to a different integration algorithm (backward euler) to
prevent this oscillation when you near steady state. Unfortunately,
Berkeley Spice3 doesn't seem to do this, and so you may notice some
sawing back-and-forth behavior near steady state. This leads to Spice3
taking more time steps than it really needs to.

Gear's method provides a bit more "numerical damping" than
trapezoidal integration, and so it avoids this problem. If you're
simulating an LC circuit, think of Gear's method as adding some small
resistors to the circuit to help damp out (false numerical) ringing.

Gear is slightly slower than trapezoidal. But if you're using a
Berkeley Spice3-derived simulator, my experience has been that Gear's
method is often a better choice than trapezoidal integration.

You can read a good layperson's discussion about the differences
between Gear and trapezoidal in Ken Kundert's book (sorry, I forget
the exact title) about Spice and Spectre.

--Eric

--
J. Eric Bracken, Ph.D.                          Tel: 1.412.261.3200 x135
Group Leader, Signal Integrity R&D              Fax: 1.412.471.9427
Ansoft Corp., Four Station Square, Suite 660    bracken@ansoft.com
Pittsburgh, PA USA 15219-1119                   http://www.ansoft.com