The long-term confusion on DC analysis

[ddrr]

spectre jacobian nonisolated

Hi, all

:?::?::?:
When we do DC analysis to a topology which contains a start-up circuit,
does the solving procedure have nothing to do with the time-domain?

thanks very much for ur answer!

 

[erikl]

When we do DC analysis to a topology which contains a start-up circuit,
does the solving procedure have nothing to do with the time-domain?

It has nothing to do with the time-domain. However, when a start-up circuit is used,
the circuit usually has 2 or more possible DC solutions, so-called non-isolated solutions (and the start-up circuit should force the desired one). Newton's method, which is used for the DC solution, however requires that the solution be isolated, and could fail with an obscure complaint about the matrix or the Jacobian being singular *). In this case the desired solution should be "encouraged" by a nodeset statement, i.e. "forcing" one or more net(s) to an expected voltage level. Note that a nodeset statement creates only a start condition and will not affect the final operating point. This is also valid for a subsequent transient simulation.

*) this has been cited from Ken Kundert's book "The Designers Guide to SPICE & SPECTRE", Chap. 2.2.3.1 "Isolated Solutions".

 

[ddrr]

It has nothing to do with the time-domain. However, when a start-up circuit is used,
the circuit usually has 2 or more possible DC solutions, so-called non-isolated solutions (and the start-up circuit should force the desired one). Newton's method, which is used for the DC solution, however requires that the solution be isolated, and could fail with an obscure complaint about the matrix or the Jacobian being singular *). In this case the desired solution should be "encouraged" by a nodeset statement, i.e. "forcing" one or more net(s) to an expected voltage level. Note that a nodeset statement creates only a start condition and will not affect the final operating point. This is also valid for a subsequent transient simulation.

*) this has been cited from Ken Kundert's book "The Designers Guide to SPICE & SPECTRE", Chap. 2.2.3.1 "Isolated Solutions".

Your answer is to the point and very helpful! Thank u!

 

[rfsystem]

So search for ALL DC solutions of a circuit is an actual research theme.


I have used a script solution to increase to chance to find others which are possible stable but not intended. It is an investigantion point for every startup circuit or meshed loops which nonlinear elements.

Script method is is to figure out all non-supply/source nodes and set them in a 2^#nodes sheme alternative to supply or ground in the nodeset. Then filtering the nonconverging out and list DC vectors which are different.

I have seen startups after fab which hangs if some ramping sequence is on the supply! But this one was only identified in simulation by changing the lower ramp value.