Deep learning approach to predict optical attenuation in additively manufactured planar waveguides

Area of Science:

• Photonics and Optical Engineering
• Machine Learning Applications
• Optical Network Design

Background:

• Growing demand for scalable optical networks requires efficient integration of fiber optics and miniaturized photonic components.
• Accurate prediction of optical losses in waveguides is crucial for optimizing network performance.
• Traditional methods for assessing waveguide attenuation are computationally intensive.

Purpose of the Study:

• To introduce a convolutional neural network (CNN) for approximating the nonlinear attenuation function of multimode waveguides.
• To develop a computationally efficient method for estimating optical losses in waveguides.
• To validate the CNN model's performance against experimental measurements.

Main Methods:

• A ray tracing model was used to simulate flexographically printed waveguide configurations and generate a dataset.
• A convolutional neural network (CNN) was trained on this dataset to learn the attenuation function.
• The trained CNN model was evaluated on test data and subsequently on fabricated waveguides.

Main Results:

• The CNN model accurately estimated optical losses due to waveguide curvature, with a standard deviation of 1.5 dB over a 50 dB range.
• The CNN model achieved a significantly faster evaluation speed (517 µs per waveguide) compared to the ray tracing model (5-10 min).
• The model demonstrated excellent agreement with experimental measurements of fabricated waveguides.

Conclusions:

• Machine learning, specifically CNNs, offers a transformative potential for revolutionizing optical network design.
• The developed CNN provides a highly efficient and accurate tool for assessing waveguide attenuation.
• This approach accelerates waveguide assessment, crucial for real-time optical network optimization.