Machine learning classification for field distributions of photonic modes
Machine learning classification for field distributions of photonic modes
复制标题
DOI：
10.1038/s42005-018-0060-1
复制
发表时间：
2018-09-28
影响因子：
5.5
通讯作者：
Becker, Christiane
中科院分区：
文献类型：
--
作者：
Barth, Carlo;Becker, Christiane
机器学习技术可以揭示大量数据中隐藏的结构，并有可能取代分析科学方法。光子纳米结构的电磁模拟通常会产生大量的数据，特别是在计算三维场分布时。一项优化任务，旨在增加发射器与光子纳米结构相互作用的光产量，加强对这些数据的系统分析。本文提出了一种结合有限元模拟和聚类的方法，用于识别具有大局域场能量和特定空间特性的光子模式。为了说明这一点，我们使用了光子晶体表面上量子点荧光的实验数值数据集。基于高斯混合模型的聚类可以将电场分布减少到最小的原型子集，并可以识别特征空间模式轮廓。提出的聚类方法有可能使生物传感、生物成像和光子上转换应用的纳米结构系统优化。
Machine learning techniques can reveal hidden structures in large amounts of data and have the potential to replace analytical scientific methods. Electromagnetic simulations of photonic nanostructures often produce data in significant amounts, particularly when three-dimensional field distributions are calculated. An optimisation task, aiming at increased light yield from emitters interacting with photonic nanostructures, enforces systematic analysis of these data. Here we present a method that combines finite element simulations and clustering for the identification of photonic modes with large local field energies and specific spatial properties. For illustration, we use an experimental-numerical data set of quantum dot fluorescence on a photonic crystal surface. The application of Gaussian mixture model-based clustering allows to reduce the electric field distributions to a minimal subset of prototypes and the identification of characteristic spatial mode profiles. The presented clustering method potentially enables systematic optimisation of nanostructures for biosensing, bioimaging, and photon upconversion applications.