AI-enhanced multislot optical gas sensors

This work presents a deep learning approach using a pre-trained deep learning model, ResNet-18, to predict the effective refractive index (neff) corresponding to different NO2 concentrations detected by MMI-based multi-slot optical gas sensors. The device is designed using the EME method to operate at 0.633 μm. The dataset is generated using the Ansys Lumerical EME solver as cross-sectional patterns of the multi-slot device with the corresponding neff. The average mean squared error (MSE) function is selected as the loss function, and the model is optimized until achieving 2.023 × 10-5 using the Adam optimizer with a learning rate of 1 × 10-4.