Taylor Robertson
Ansys Employee

I would be careful with 2D simulations is this varFDTD or 2D FDTD? There are separate approximations for each polarization in both cases. Even if the sources are the same polarization, other difference like the normalization could changing this. See Understanding Frequency Domain Normalization.

Usually you would validate and reoptimize any topology optimized devices in 3D to avoid breaking the 2D assumptions. It is likely the performance will change in 3D, but the built in 2D assumptions mean that the collective results will be more consistent.