Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!swrinde!cs.utexas.edu!uunet!cadence!steveg
From: steveg@cadence.com (Steven Greenberg; x6231)
Newsgroups: comp.lsi.cad
Subject: Re: Modeling of ideal diodes in (H)spice
Keywords: simulation, spice, modeling, convergence
Message-ID: <1991Mar19.185326.4818@cadence.com>
Date: 19 Mar 91 18:53:26 GMT
References: <1991Mar16.211609.9149@zip.eecs.umich.edu>
Sender: usenet@cadence.com (USENET News)
Organization: Cadence Design Systems, Inc.
Lines: 43

In article <1991Mar16.211609.9149@zip.eecs.umich.edu> ayman@eecs.umich.edu writes:
>I am trying to simulate a circuit in Hspice, where all diodes are modeled by
>a switched voltage source in series with a resistor.
>
>        _______ \__________/\/\/\/\/\______
>        +                     Rd           |              switches closes when
>                                           |              Vd becomes greater
>        Vd                             ----+----  Vi      than Vi
>                                          ===
>        -                                  |
>        -----------------------------------+
>
>I tried the voltage-controlled resistor switch (Rd = 1M for Vd < Vi, Rd = 100
>for Vd > Vi), but the dc solution didn't converge.
>Any suggestions on how to make the solution converge? Or, are there other ways
>to model the diodes to get a convergent solution?
>Thanks.
>
>--------------------------------------------
>- Ayman I. Kayssi (ayman@eecs.umich.edu)   -
>- Advanced Computer Architecture Lab.      -
>- EECS Dept., Univ. of Michigan, Ann Arbor -
>--------------------------------------------


Piece-wise linear circuits like this (especially with very coarse
pieces) are not too friendly to Spice's Newton-Raphson non-linear
equation solving algorithm.  The problem is that the resistance in one
segment is not a good predictor of the behavior of the component when
it switches segments.  Newton-Raphson depends on the predictive nature
of the derivative (resistance).

You might need to give Spice a large number of iterations to find the
answer.  Unfortunately, piece-wise-linear circuits can go into
oscillations in the Newton-Raphson solver so that no amount of iterations
would be sufficient.

By the way, the switch as you have described it has about a microamp
of negative current at zero volts.  You might want to add a parallel
current source to cancel out this current.  This current may or may not
matter in your circuit.

/Steve