Learning Forum

[Andy Li]

Hello,

I'm looking to optimize a device, using Lumopt's topology optimization, that is intended to operate at two different temperatures. I tried doing so by defining two different geometries (TopologyOptimization2D objects) in the Lumopt script that have different refractive indices for the background and waveguide materials. In addition, I added separate optimizations with different refractive indices for the background and waveguides.

However, the resulting device had very poor simulated performance which was significantly worse than a device that had not been optimized for temperature at all. In addition, I noticed that the final value of the refractive index for the higher temperature (and higher refractive index) device was lower than the refractive index implied by the waveguide permittivity set in the geometry definition.

 

Is there a working method to optimize a device for more than one temperature in Lumopt?

[Devika]

Hello Andy, 

Apologis for the delay in response. Could you please share the optimization progess? 

There is a chanace the optimization didnt run to completion. 

Are you optimizing both temperature simultaniously (superopt:Getting Started with lumopt - Python API – Ansys Optics)?

 

I am sharing a link might here: Optimizable Geometry - Python API – Ansys Optics

Thanks 

Devika

[Andy Li]

Hello Devika,

I am optimizing for both temperatures simultaneously. The optimization has finished running; the end device is about 99% binarized.

Thanks,

Andy

[Taylor Robertson]

Hello Andy,

Could you share the optimization progress, and compare this with the performance of the unoptimized device? I don't understand how the device could have worse performance then the initial design? The optimizer could certainly get stuck in a local minima, but perhaps starting from your best candidate then proceeding with lumopt at the both temperatures could help?

Also how are you assigning the defining the device and background index via python. Could you share some code snippets? Are you using varFDTD?

Thanks,

 

[Andy Li]

Hello,

I tried starting from the best candidate (which was optimized at one temperature) but the same issue occurred - the performance at both temperatures worsened.

I'm defining the device with an LSF file and editing it for one optimization:

The optimized geometry is set using the same parameter matrix but different values for eps:

This is being optimized using 2D FDTD.

[Taylor Robertson]

In superopt it optimizes the toal FOM, so one FOM could reduce if the other improves by a greater amount; however, it shouln't be possible for the performance to reduce overall. The optimization wouldn't proceed in that case, unless I am missing something it seems that there may be an issue with the simulation set-up or opt definition.

Maybe double check that the eps values are being set correctly, by looking at the index monitors?

Can you share the optimization progress report.

 

I would also note that the performance between 2D and 3D could be quite significant. While varFDTD tries to make 2D FDTD more accurate, the assumptions may not be valid in topology.

 

Thank you,

[Andy Li]

Hello,

Here's the progress of the FOM in the optimization:

I've also checked that the eps values are set correctly by looking at the index monitors on the forward-optimization FSP files.

[Taylor Robertson]

Hello Andy,

 

This is plotting the error from the metric function, so it appears that it has improved a reasonable amount in the first 20 iterations. Is it clear how you should define the figure of merit? After that it seems the performance for FOM_1 and FOM_2 gets better, but 3/4 gets worse, the total improves very marginally. It then seems that binarization starts, but the performance doesn't get significantly worse. It looks like you could continue running. If required.

 

Thanks,

[Andy Li]

Hello Taylor,

I noticed that all of the figures of merit seem to trend downward (improving), yet I still notice significantly worse performance. Could it be due to some sort of scaling of the figure of merit?

 

Thanks,

Andy

[Taylor Robertson]

Hello Andy,

 

It seems likely that the FOM definition doesn't match up with how you considering performance. Maybe you copuld share some code snippets from that section and I will see if I can see any issue? See ModeMatch - https://optics.ansys.com/hc/en-us/articles/360050995394.