High-Accuracy Prediction of Metal-Insulator-Metal Metasurface with Deep Learning

Kaizhu Liu,Hsiang-Chen Chui,Changsen Sun,Xue Han

2023-08-04

Abstract:Deep learning prediction of electromagnetic software calculation results has been a widely discussed issue in recent years. But the prediction accuracy was still one of the challenges to be solved. In this work, we proposed that the ResNets-10 model was used for predicting plasmonic metasurface S11 parameters. The two-stage training was performed by the k-fold cross-validation and small learning rate. After the training was completed, the prediction loss for aluminum, gold, and silver metal-insulator-metal metasurfaces was -48.45, -46.47, and -35.54, respectively. Due to the ultralow error value, the proposed network can replace the traditional electromagnetic computing method for calculation within a certain structural range. Besides, this network can finish the training process less than 1,100 epochs. This means that the network training process can effectively lower the design process time. The ResNets-10 model we proposed can also be used to design meta-diffractive devices and biosensors, thereby reducing the time required for the calculation process. The ultralow error of the network indicates that this work contributes to the development of future artificial intelligence electromagnetic computing software.

Machine Learning,Computational Physics,Optics

What problem does this paper attempt to address?

The paper aims to address the high-precision prediction problem of the S11 parameter of metal-insulator-metal (MIM) metasurfaces. Specifically, the authors propose a method based on the ResNets-10 model to predict the S11 parameters of MIM metasurfaces composed of different metal materials (aluminum, gold, and silver). The core challenge of the study is to improve prediction accuracy to replace traditional electromagnetic software calculation methods.

The main contributions of the paper include:
1. **High-precision prediction**: Through two-stage training and k-fold cross-validation, extremely low prediction errors were achieved. The prediction losses for aluminum, gold, and silver MIM metasurfaces were -48.45, -46.47, and -35.54, respectively.
2. **Fast training**: Compared to existing literature, the model requires less than 1,100 epochs to complete training, significantly reducing the time needed for the design process.
3. **Transfer learning application**: By using transfer learning techniques, the prediction performance for different materials was improved, further enhancing the model's generalization ability.
4. **Practical application**: The model can be used for designing meta-diffraction devices (such as metalenses and gratings) and surface-enhanced Raman scattering (SERS) applications, greatly reducing the trial-and-error time and cost required by traditional calculation methods.

In summary, the paper proposes a new deep learning framework to solve the efficient and accurate prediction problem of the S11 parameter of MIM metasurfaces and demonstrates its potential in various practical application scenarios.