Question about DC convergence error in sweep simulation.

Hi Dongsuk,

It's extraordinarily rare for Spectre to produce an "incorrect" DC operating point (so rare, I can't remembering it ever happening). It's possible however that the operating point is perfectly valid, but not the one you want - this might be because of:

1. The circuit having multiple operating points
2. The device models (including Verilog-A models) having non-physical behaviour which means that an alternative non-physically correct operating point is a valid solution

For the first, this is a function of your circuit - and normally you'd want to know about that; it might be expected, or it might be undesirable. For the second, you may need to ensure that your model actually makes sense (for example, I've seen major foundries model non-linear resistors in Verilog-A and forgetting about the fact that their quadratic equation for the resistance has two roots, one of which is a completely unreasonable answer electrically).

Usually if you can capture the nodeset and use it (e.g. by doing your sweep and then using writefinal to capture the last point in the sweep, and using that as a nodeset with ever other condition the same as that point) the results should be identical in that case. That's because if that solution was valid at that point, then there should be no need to iterate away from it. Of course, you would want to be very sure that the results are reasonable and that there isn't another perfectly reasonable operating point that could occur in the real circuit.

It's hard to give a more specific response without seeing the actual case - so you might want to contact customer support and log a support case.

Andrew

Hi Dongsuk,

It's extraordinarily rare for Spectre to produce an "incorrect" DC operating point (so rare, I can't remembering it ever happening). It's possible however that the operating point is perfectly valid, but not the one you want - this might be because of:

1. The circuit having multiple operating points
2. The device models (including Verilog-A models) having non-physical behaviour which means that an alternative non-physically correct operating point is a valid solution

For the first, this is a function of your circuit - and normally you'd want to know about that; it might be expected, or it might be undesirable. For the second, you may need to ensure that your model actually makes sense (for example, I've seen major foundries model non-linear resistors in Verilog-A and forgetting about the fact that their quadratic equation for the resistance has two roots, one of which is a completely unreasonable answer electrically).

Usually if you can capture the nodeset and use it (e.g. by doing your sweep and then using writefinal to capture the last point in the sweep, and using that as a nodeset with ever other condition the same as that point) the results should be identical in that case. That's because if that solution was valid at that point, then there should be no need to iterate away from it. Of course, you would want to be very sure that the results are reasonable and that there isn't another perfectly reasonable operating point that could occur in the real circuit.

It's hard to give a more specific response without seeing the actual case - so you might want to contact customer support and log a support case.

Andrew

Dear Andrew,

First of all, thank you so much for the kind answer. I tried to do what you said so my answer has been little bit late. My apologize.

It seems like licence key is needed to use customer support ... I need to ask school department about it.

However, about what you suggested, I saved sweep value via 'writefinal' and used this in AC simulation 'readfile' option, and until now it seems it works.

Maybe for sturdy design I should look inside the verilog-A code ... maybe there are some problem when I'm modulating it.

Again, thank you so much for helping me.

Regards,

Dongsuk