I try to optimize a layout using the SONNET - interface. If I perform a "normal" analysis, everything works fine.
If I set up the optimization goals, these are shown in the result graph window as desired.
If I start the optimization, SONNET ist called one time and the result of the error function ist shown. But then nothing more happens, except that the the "round counter" is increased very fast, without changing the error function any more. I guess that even the variables aren´t changed...
Shouldn´t the optimizer change a variable, call SONNET, calculate the error, change variables and so on?
Does Anybody have any hints? Or is there even an example covering this topic?
Thanks a lot,
Matthias
If I set up the optimization goals, these are shown in the result graph window as desired.
If I start the optimization, SONNET ist called one time and the result of the error function ist shown. But then nothing more happens, except that the the "round counter" is increased very fast, without changing the error function any more. I guess that even the variables aren´t changed...
Shouldn´t the optimizer change a variable, call SONNET, calculate the error, change variables and so on?
Does Anybody have any hints? Or is there even an example covering this topic?
Thanks a lot,
Matthias
Genesys does not optimize the geometry of electromagnetic simulations (although this feature is planned for a future release). Currently, you can run an optimization on an empower or sonnet simulation, but it will only optimize the co-simulation part of the simulation. That is, it will only change the values of lumped elements in the layout that have been made tunable in the schematic. If you need to optimize a part of the layout besides lumped elements the best method is to optimize it using the linear simulation of the schematic and then run the sonnet simulation on the new layout. Be sure to check your layout before you run the sonnet simulation. You may need to reposition parts of it depending on how the optimization changed your structure.
An additional technique that is commonly used at Eagleware is to tune (or optimize) the linear to match the electromagnetic simulation, then reverse the changes on the original circuit. A simple example: If the EM simulation has a notch (due to a stub) at 1.2 GHz, but the linear simulation predicts the notch at 1.25 GHz, lengthen the stub in the linear simulation until it matches the EM simulation (maybe 50 mils longer). Then, going back to the original circuit, REVERSE the change by making the stub 50 mils SHORTER. This technique can be applied easily to more complicated circuits (sometimes using optimization), and it typically works MUCH better and faster than true electromagnetic optimization.
Ryan