A Machine Learning Framework for Quantum Cascade Laser Design

TL;DR

This paper presents a machine learning framework using neural networks to rapidly predict and optimize quantum cascade laser designs, significantly reducing computational time while improving the figure of merit.

Contribution

The authors developed a neural network-based method to predict laser performance and optimize designs, achieving faster results and better laser structures compared to traditional simulations.

Findings

Neural network predictions show 5-15% error for potential laser structures.

The method identified high-performance laser designs with a 1.5-fold increase in figure of merit.

Computational time was reduced from 32 hours to 8 hours for analyzing 907 million designs.

Abstract

A multi-layer perceptron neural network was used to predict the laser transition figure of merit, a measure of the laser threshold gain, of over 900 million Quantum Cascade Laser designs using only layer thicknesses and the applied electric field as inputs. Designs were generated by randomly altering the layer thicknesses of an initial 10-layer design. Validating the predictions with our 1D Schr\"odinger solver, the predicted values show 5% to 15% error for structures where a laser transition could occur, and 35% to 70% error for structures where there was no laser transition. The algorithm allowed (i) for the identification of high figure of merit structures, (ii) recognition of which layers should be altered to maximize the figure of merit at a given electric field, and (iii) increased the original design figure of merit of 94.7 to 141.2 eV ps \r{A}^2, a 1.5-fold improvement and significant for QC lasers. The computational time for laser design data collection is greatly reduced from 32 hours for 27000 designs using our 1D Schr\"odinger solver on a virtual machine, to 8 hours for 907 million designs using the machine learning algorithm on a laptop computer.