计算机科学
级联
反向
量子
激光器
超晶格
过程(计算)
测距
量子级联激光器
拓扑(电路)
计算机工程
电子工程
光电子学
电气工程
材料科学
光学
数学
工程类
电信
物理
操作系统
量子力学
化学工程
几何学
作者
Y. Hu,Suraj Suri,Jeremy Daniel Kirch,Benjamin B. Knipfer,Zongfu Yu,D. Botez,Luke James Mawst
出处
期刊:IEEE Photonics Conference
日期:2022-11-01
卷期号:: 1-2
被引量:3
标识
DOI:10.1109/ipc53466.2022.9975491
摘要
The active core (AC) of a quantum cascade laser (QCL) is a superlattice with hundreds of layers, which embodies stages consisting of 10-12 quantum wells repeating 30 to 50 times. Optimization of these layer thicknesses and compositions yields the desired device performance metrics for a given structure. As part of the device-design processfor these structures, human intervention is needed at multiple steps from identifying relevant wavefunctions to analyzing device-performance metrics. This is a time-consuming process ranging from weeks to months with little to no guarantee of success. We aim to replace this human intervention aspect of the AC design by training deep neural networks (DNNs) and implement an inverse design approach. To train these DNNs, datasets of QCL-AC structure and corresponding performance metrics are used to develop instinctive connections between them. After training is finished, we expect this network to achieve an AC design based on input device performance metrics in a significantly shorter timeframe.
科研通智能强力驱动
Strongly Powered by AbleSci AI