研究目的
To design photonic crystal based optical filters using a machine learning based mathematical model to predict the spectral response and reduce simulation time and efforts.
研究成果
The proposed machine learning based mathematical model effectively predicts the spectral response of dielectric photonic crystal optical filters, reducing design time and simulation efforts. The linear regression equation accurately models wavelength shifts, while polynomial regression handles non-linear transmission behaviors. Future work should include more parameters and data for enhanced modeling.
研究不足
The model is simplified, considering only two design parameters (hole radius and waveguide thickness) and a single resonant mode. It may not account for all physical properties or multiple resonant modes. The training data is limited to specific ranges, and higher-order regressions or more data could improve accuracy.
1：Experimental Design and Method Selection:
The study uses a finite-difference time-domain (FDTD) method with MEEP software for numerical simulations of photonic crystal structures. Machine learning algorithms are applied to train a mathematical model based on simulation data.
2：Sample Selection and Data Sources:
A dataset of over 170 different sets of design parameters (hole radius and waveguide thickness) is generated through simulations, with parameters varied in specific ranges.
3：List of Experimental Equipment and Materials:
MEEP software (open-source FDTD simulator), computational resources for simulations. Materials include ZrO2 or NbO5 for waveguide (refractive index 2.2) and SiO2 for substrate and cladding (refractive index 1.5).
4：2) and SiO2 for substrate and cladding (refractive index 5).
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Design the PhC structure with cylindrical air-holes in a waveguide layer. Use single-cell model with periodic boundary conditions and perfectly matched layers. Simulate with a plane wave source over a wavelength range of 1.9 μm to 2.2 μm. Extract transmission spectra and train machine learning models (linear and polynomial regression) to derive equations relating design parameters to spectral responses.
5：9 μm to 2 μm. Extract transmission spectra and train machine learning models (linear and polynomial regression) to derive equations relating design parameters to spectral responses.
Data Analysis Methods:
5. Data Analysis Methods: Analyze transmission spectra to extract central wavelength and transmission-dip values. Use machine learning for regression modeling, comparing simulation results with model predictions.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
