Yujia YuanPhysical systems · Stanford EE

Home / Projects / P-04

Algorithm for photonic pattern design

ScaleOther
PlaceWaterloo · IQC · Bajcsy Lab
FieldComputational photonics
Period2020 — 2021
Fig. 1 — Inverse-design geometry before / after fabricability smoothing.
Fig. 1 — Inverse-design geometry before / after fabricability smoothing.
Fig. 2 — Second inverse-design fabricability example.
Fig. 2 — Second inverse-design fabricability example.

Nowadays, more and more photonics groups rely on AI to generate photonic designs that have ideal behavior. This approach might produce mathematically strong designs that are physically annoying to fabricate: sub-resolution holes, fragile bridges, and sharp edges that will round or fail during processing.

We implemented a computer-vision algorithm that removed unwanted regions and made sure the resulting design is fully fabricable. After the algorithm, the generated pattern was simulated again to ensure that FoM decreased by less than 5%. Designs that do not preserve FoM are re-generated.

Result. AI generated photonics designs can first be reliably produced in real life. Those AI tools are no longer academic toys but productivity.