Enterprise AI Analysis: Optics & Photonics

Balancing FOM and Bandwidth: Optimizing 2D-material-based optical polarizers involves a critical trade-off between maximizing the Figure of Merit (FOM) and achieving a sufficient operation bandwidth. Our analysis shows that the highest FOM values are often achieved near the mode convergence boundary, but this can limit the device's operational bandwidth. The FCNN model is particularly powerful in rapidly mapping the global variation trend of FOM' across the entire parameter space. This capability provides invaluable guidance for designers to choose optimal (W, H) parameters that balance high FOM with required bandwidth for specific applications, ranging from narrow-band optical communications (~100 nm) to broader optical sensing (hundreds of nm).

Training Data & Accuracy Trade-offs: The FCNN model's prediction accuracy is influenced by the size and resolution of the training dataset. As the step size (Δ) between adjacent structural parameters decreases, the training dataset grows, generally leading to increased accuracy. However, this also significantly increases the computational cost of generating the training data. For example, reducing Δ from 80 nm to 20 nm increased the dataset size by approximately 14.7 times. Our studies indicate that a Δ = 20 nm offers the best balance, providing high accuracy (e.g., ~99.0% for FCNN-1 and AD of ~0.018 for FCNN-2) without incurring excessively high training data generation costs. This flexibility allows tailoring the dataset construction strategy to balance efficiency and accuracy based on specific application requirements.