Machine learning helps to improve photonic applications



[ad_1]

The computer simulation shows how the electromagnetic field is distributed in the silicon layer with a pattern of holes after excitation with a laser. Here, bands with local field maxima are formed, so that the quantum dots shine particularly brightly. Credit: Carlo Barth / HZB

In addition to solar cells, photonic nanostructures can be used for many applications, for example optical sensors for cancer markers or other biomolecules. An HZB team using computer simulations and machine learning has now shown that the design of such nanostructures can be optimized selectively. The results are published in Physics of communications.

Nanostructures can significantly increase the sensitivity of optical sensors, provided that the geometry meets certain conditions and corresponds to the wavelength of the incident light. Indeed, the electromagnetic field of the light can be greatly amplified or reduced by the local nanostructure. The group of young researchers HZB "Nano-SIPPE" led by Professor Christiane Becker works on the development of this type of nanostructures. Computer simulations are an important tool for this. Carlo Barth, of the Nano-SIPPE team, has now identified the main models of field distribution in a nanostructure using machine learning and explained the experimental results.

The photonic nanostructures examined in the paper consist of a silicon layer with a regular hole pattern covered with lead sulfide quantum dots. Excited with a laser, the quantum dots close to local field amplifications emit much more light than on an unordered surface. This demonstrates empirically how laser light interacts with the nanostructure.

In order to record what happens when the individual parameters of the nanostructure change, Barth calculates the three-dimensional distribution of the electric field for each set of parameters with the help of software developed at the Zuse Institute Berlin. Barth has analyzed these huge amounts of data with other computer programs based on machine learning. "The computer has examined about 45,000 records of data and has grouped them into a dozen different models," he says. Finally, Barth and Becker have identified three basic models in which fields are amplified in specific areas of nanoholes.

This optimizes the photonic crystal membranes according to the excitation amplification for virtually any application. Some biomolecules preferentially accumulate along the edges of the hole, for example, while others prefer trays between holes, depending on the application. With correct geometry and correct excitation of light, the maximum amplification of the electric field can be generated exactly at the binding sites of the desired molecules. This would increase the sensitivity of optical sensors for cancer markers at the level of individual molecules, for example.


Explore more:
Patented nanostructure for solar cells: coarse optics, smooth surface

More information:
Carlo Barth et al., Machine Learning Classification for Photonic Mode Field Distribution, Physics of communications (2018). DOI: 10.1038 / s42005-018-0060-1

Provided by:
Helmholtz Association of German Research Centers

[ad_2]
Source link