iabstol, lowering of

[Roger Bourne]

Hello all,

I have a cct (fairly simple and small) that I am simulating in
Spectre. I can push the simulation parameter iabstol to 1e-15, but no
less. When, I set iabstol to 1e-16 or less, my DC, tran and noise
simulations have convergence problems. Unfortunately, I need a better
accuracy than that, in the neighborhood of 50e-18A (or less).
How can the simulator to run with an iabstol smaller than 1e-15?


Other simulator parameters are :
reltol= 1e-6
vabstol= 1e-6
gmin=1e-18
The first 2 parameters have been varied, but alas, these variations do
not permit the simulator to accept an iabstol of less than 1e-15.

Any help will be appreciated
Thank you.

Regards,
-Roger

 

[Andrew Beckett]

On 23 Mar 2007 12:16:38 -0700, "Roger Bourne" <rover8898@hotmail.com> wrote:

Hello all,

I have a cct (fairly simple and small) that I am simulating in
Spectre. I can push the simulation parameter iabstol to 1e-15, but no
less. When, I set iabstol to 1e-16 or less, my DC, tran and noise
simulations have convergence problems. Unfortunately, I need a better
accuracy than that, in the neighborhood of 50e-18A (or less).
How can the simulator to run with an iabstol smaller than 1e-15?


Other simulator parameters are :
reltol= 1e-6
vabstol= 1e-6
gmin=1e-18
The first 2 parameters have been varied, but alas, these variations do
not permit the simulator to accept an iabstol of less than 1e-15.

Any help will be appreciated
Thank you.

Regards,
-Roger

How big are the currents and voltages in the circuit? Normally the abstols are
set to about 1 millionth of a typical signal level in the circuit. If you over
tighten them, then you can run out of numerical headroom (double precision
numbers have about 15 digits of accuracy - but you need some margin for the
iterative numerical algorithms). A reltol of 1e-6 is quite tight too, and is
not that commonly needed...

Perhaps you can explain _why_ you need such a tight iabstol?

Regards,

Andrew.
--
Andrew Beckett
Principal European Technology Leader
Cadence Design Systems, UK.