[17p-K305-1~12] 3.13 Silicon photonics, Photonics-electronics convergence, Optical control

2:15 PM - 2:30 PM JST(5:15 AM - 5:30 AM UTC)

[17p-K305-4] Realization of high-speed 3D waveguide analysis model via transfer learning based on 2D-FDTD simulation

〇(B)Gai Ichisawa¹, Sho Okada², Tomohiro Amemiya¹ (1.Science Tokyo, 2.NICT)

Keywords:

silicon photonics,simulation,machine learning

To reduce the high computational cost of 3D simulation, we propose a method that trains a neural network using 2D FDTD data and applies transfer learning with a small amount of 3D FDTD data, enabling predictions equivalent to 3D FDTD results at low cost and in a short time. Using an MMI device as an example, the proposed method achieved a coefficient of determination of R² = 0.91 for 2D FDTD and R² = 0.81 for 3D FDTD, demonstrating high prediction accuracy.