Characterization of novel gold nanoparticles throughoptical spectroscopy and convolutional neural networks

Plasmonic nanoparticles exhibit large optical cross-sections that have attracted much interest for applications in photocatalysis, imaging, and nano-optics. However, the plasmonic behavior in these applications depends on particle properties such as relative proximity, crystallinity, size, and shape. The most common particle fabrication techniques, colloidal synthesis and electron beam lithography, allow to tailor these properties to an extent, yet typically result in particle-to-particle variations that broaden the plasmonic properties. Correlated single-particle approaches are required to resolve the governing mechanisms, but can be experimentally difficult and tedious. Here, we utilize polymer pen lithography to synthesize gold nanoparticles on glass and propose a strategy to characterize both their optical and structural properties through optical spectroscopy. We find that polymer pen lithography yields organized arrays of crystalline nanoparticles with controlled particle size, precise relative position, and high throughput. We take advantage of this synthesis technique by systematically measuring optical scattering spectra using an automated microscope. Through correlation with electron microscopy images, we build a convolutional neural network that predicts structural information from optical scattering. Our work suggests polymer pen lithography as a promising synthesis approach for plasmonic nanoparticles and offers an efficient way to study their properties.