Hi,

 Recently while I use spectre6.1.1 to simulate an osc with PSP model, I found the results can vary in a very strange fashion. My testbench is like following:

1. I use "width" as a variable for transistor width of which the default value is 10 (um).
2. Use parametric to sweep the width from 6 to 16 with step of 2.
3. I found the results of "width=10" are different in 1. and 2. And the difference can be up to 2 dB.


I did make sure the osc has enough time to settle, and set the accuracy requirement to 1e-6 relative, and 1e-8 absolute. I also check the netlist directly, but cannot find any discrepency.

Does anybody have any clue this kind of problem could happen? Maybe this is due to the computer or other things rather tahn simulator?



Sincerely,

It's quite possible (especially with an oscillator) to have a sensitive dependency on the initial conditions. When you're doing a sweep, the dc solution will start from the previous solution.

You might want to ensure that the transient or pss analysis you're using has "restart=yes", and also set the simulation option "rebuildmatrix=yes" (I think that's the parameter - I don't have access to spectre at the moment). "spectre -h options" will tell you what it's called.

Regards,

Andrew,

Ah, close but no underscore...

Thanks Scott for correcting my memory!

Regards,

Andrew.