Neural-Network-Enabled Design of a Chiral Plasmonic Nanodimer for Target-Specific Chirality Sensing.

Quantitative analysis of chiral molecules in various solvents is essential. However, there are still many challenges to enhancing the sensitivity in precisely determining both concentration and chirality. Here, we built an algorithmic methodology to predict and optimally design the chiroptical response of chiral plasmonic sensors for a specific target chiral analyte with the aid of deep learning. Based upon the analytic and intuitive understanding of the Born-Kuhn type plasmonic nanodimer, we designed and trained the neural networks that can successfully predict the chiroptical properties and further inversely design the plasmonic structure to achieve the intended circular dichroism. The developed algorithm could identify the optimum structure exhibiting the maximum sensitivity for the given specific analytes. Surprisingly, we discovered that sensitivity strongly depends on the various conditions of analytes and can be finely tuned with the structural parameters of plasmonic nanodimers. We envision that this study can provide a general platform to develop ultrasensitive chiral plasmonic sensors whose structure and sensitivity have been evolved algorithmically for adoption in specific applications.