Predictions of resonant mode characteristics for terahertz quantum cascade lasers with distributed feedback utilizing...

Area of Science:

• Optoelectronics and Photonics
• Applied Physics
• Machine Learning in Engineering

Background:

• Terahertz quantum cascade lasers (THz QCLs) are leading solid-state THz sources, with distributed feedback (DFB) gratings enabling single-mode emission and high performance.
• Accurate prediction of THz QCL resonant mode characteristics (frequency, loss, electric-field distribution) is crucial for device analysis and fabrication.
• Traditional numerical simulations for these characteristics are computationally intensive, time-consuming, and involve accuracy-efficiency trade-offs.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting resonant mode characteristics of THz QCLs with various DFB structures.
• To provide a highly accurate and efficient alternative to conventional numerical simulation methods for THz QCL design.
• To demonstrate the broad applicability of the developed ML models across different DFB configurations and advanced THz QCL designs.

Main Methods:

• Development of ML models using a multi-layer perceptron for frequency and loss prediction, and an up-sampling convolutional neural network for electric-field distribution prediction.
• Training and validation of models using over 1000 samples based on four key structural parameters: grating period, waveguide length, grating duty cycle, and highly-doped contact layer length.
• Evaluation of model performance using Pearson correlation coefficients and peak signal-to-noise ratios, alongside prediction time analysis.

Main Results:

• ML models achieved high accuracy, with Pearson correlation coefficients exceeding 0.99 for frequency and loss predictions.
• Electric-field distribution predictions demonstrated high fidelity, with median peak signal-to-noise ratios above 33.74 dB.
• The developed ML models provide predictions within seconds, significantly outperforming traditional simulations in efficiency.

Conclusions:

• The proposed ML models offer a highly accurate and efficient approach for predicting resonant mode characteristics of THz QCLs with DFB gratings.
• These models are widely applicable to diverse DFB structures and can accurately predict characteristics for advanced resonators like graded photonic heterostructures and phase-locked arrays.
• The ML-driven approach accelerates THz QCL research and development by reducing computational burden and enabling rapid design exploration.