For most of us, color forms an integral and irreplaceable part of our lives from a very early age. For centuries, humanity has artificially induced color using dyes, pigments, and, more recently, pixels, but many of the vibrant colors seen in nature are created on a smaller and more integral scale.
Unlike man-made pigments that absorb certain wavelengths of light and reflect others, the intricate and beautiful patterns of a butterfly’s wings or a peacock’s feathers are created when light interacts with tiny nanostructures, thus creating vibrant and often iridescent colors.
This arises as the result of interactions of light and tiny collections of holes at the nanometer scale called nanohole arrays that have the unique ability to manipulate light to produce what is known as “structural color”.
If structural color can be employed in man-made materials like it is in the butterfly’s wings, this could give rise to a more vibrant and longer-lasting color that doesn’t degrade over time. Because this is a physical rather than a chemical coloration, the process could do away with the need for dyes that can be environmentally unfriendly or harmful to human health. Thus, scientists have been diligently working to replicate structural color.
One of the major setbacks in this research currently is the fact that it is difficult to create a nanohole arraysthat results in a specific color, as this requires precise control of nanoscale structure, a complex and challenging task, to say the least.
In a new paper published in the journal Advanced Intelligent Systems, a team led by Bin Ai from the School of Microelectronics and Communication Engineering at Chongqing University, Chongqing, designed a new method that could design nanohole arrays that can produce a specific structural color. This required the new turning to machine learning in their structural color quest.
“Two deep learning networks, known as color–structure–color (CSC) and color–structure–spectrum (CSS), were developed to predict the structural color of NAs based on their geometric parameters, and vice versa,” Ai said. “The research was primarily based on simulations, utilizing finite-difference time-domain calculations to generate a dataset of transmission spectra and structural parameters of [nanohole arrays].”
The future is bright for structural color
The researchers found that this inverse approach of predicting colors based on structures with CSC and CSS in the simulation allowed the creation of nanohole arrays that can create a desired color, thus opening up the possibility to reproduce colorful images on a nano-scale.
“The accuracy, function, and efficiency of the system were remarkable, showcasing the potential of combining deep learning with nanoscience,” Bin Ai said, explaining even the team behind the research was surprised with the efficiency seen in the results achieved.
While a significant step towards nanoscale structure and a testament to the practical uses of deep learning, the team points out that the research and its results are based solely on simulation.
The next step is to translate these simulation results into experimental reality and to develop a prediction model for practical experiments based on these deep-learning results. This may involve factoring in more “real-world” considerations to prediction models.
“The team plans to use transfer learning to translate simulation results into experimental reality,” Bin Ai said. “This approach will help in predicting experimental outcomes, bridging the gap between theoretical work and practical applications.”
The development of structural color has practical implications in a wide range of fields. For instance, nanohole arrays could be used to endow hard-to-replicate unique security features to banknotes that help prevent counterfeiting or could be introduced to substances to indicate the presence of particular chemicals. The nanohole array control system could also be employed for better high-density data storage.
Reference: Bin Ai, et al., Inverse Design of Plasmonic Nanohole Arrays by Combing Spectra and Structural Color in Deep Learning, Advanced Intelligent Systems (2023). DOI: 10.1002/aisy.202300121
Feature image credit: Joshua J. Cotten on Unsplash