Machine Learning Helps To Optimize Photonics Applications

Photonic nanostructures not only increase the efficiency of solar cells, but also enhance the effectiveness of optical sensors used, for example, as cancer markers. Using computer simulations and the use of machine learning, a team at HZB has now shown how the design of such nanostructures can be specifically optimized. The results are published in Communications Physics.

With nanostructures, the sensitivity of optical sensors can be increased enormously - provided that the geometry fulfills certain conditions and matches the wavelength of the incident light. Because the electromagnetic field of light can be extremely intensified or weakened locally by the nanostructure. At HZB, the junior research group Nano-SIPPE is working with Prof. Dr. med. Christiane Becker to specifically develop such nanostructures. An important tool in this case are computer simulations. Dr. Using machine learning, Carlo Barth from Beckers team has now identified the most important patterns of field distribution in a nanostructure and thus very well explains the experimental findings for the first time.

Nanostructures: Light makes quantum dots shine
The photonic nanostructures considered in this work consist of a silicon layer with a regular hole pattern coated with lead sulfide quantum dots. Excited with a laser, the quantum dots shine much stronger through the local field increases than on an unstructured surface. This allows experimentally to show how the laser light interacts with the nanostructure.

Ten different patterns
In order to systematically record what happens when individual parameters of the nanostructure change, Barth calculated the three-dimensional field distribution for each parameter set using a software developed at the Zuse Institute in Berlin. Barth then analyzed these enormous amounts of data from other computer programs based on machine learning methods. "The computer has scoured the roughly 45,000 data sets and grouped them into about ten different patterns," explains Barth. Finally, Barth and Becker succeeded in crystallizing three basic patterns among others, in which the fields are reinforced in various specific areas of the nanoholes.

Sensors for individual molecules, for example cancer markers
This now allows the optimization of photonic crystal membranes for virtually any application based on excitation enhancement. For depending on the application, some biomolecules, for example, preferentially accumulate along the edges of the holes, others more on the plateaus between the holes. With the right geometry and appropriate excitation by light, the maximum field enhancement could then be created exactly at the attachment sites of the sought-after molecules. This would increase the sensitivity of optical sensors, for example for cancer markers, to the level of single molecules.