PIC-Flow applies conditional flow matching with a real-valued U-Net and interface-masked Helmholtz residual loss to predict electromagnetic fields in photonic devices, generalizing to held-out device classes beyond its training set.
All mea- surements are taken on the same compute node: an NVIDIA A100 GPU paired with dual AMD EPYC 75F3 processors (2×32 physical cores, Zen3)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
physics.optics 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Physics-Based Flow Matching for Full-Field Prediction of Silicon Photonic Devices
PIC-Flow applies conditional flow matching with a real-valued U-Net and interface-masked Helmholtz residual loss to predict electromagnetic fields in photonic devices, generalizing to held-out device classes beyond its training set.