Reinforcement learning for plasmonic nanomaterial discovery and sensing

In fifteen words or less, explain the significance of this contribution (Novel Aspect).: Autonomous flow platform combines reinforcement learning and nanoparticle growth for real-time chemical sensing.

Abstract Text: In a research landscape increasingly shaped by artificial intelligence (AI), we aim to accelerate the
discovery and optimization of plasmonic nanomaterials for next-generation nanosensors. Surface-enhanced
Raman scattering (SERS) is a non-invasive (bio)analytical technique that detects the vibrational signatures
of molecules as they interact with a plasmonic nanostructured surface. Our aim is to develop highly
sensitive SERS sensors optimized for near-infrared (NIR) light sources, due to their greater tissue
penetration and suitability for biological analysis. This work focuses on advancing SERS-based chemical
sensing through AI-driven nanomaterial innovation.
We have optimized the production of SERS substrates, gold nanostars (AuNS) and have shown their highly
tunable morphology in a continuous flow system. This setup allows precise control over AuNS optical
responses influenced by its morphology via surface plasmon resonance (SPR), producing particles with
optical responses spanning the visible to the NIR region (600 to 1100 nm), perfect for our goal of designing
NIR SERS substrates. We validated this approach through electron microscopy, in-line and off-line UV-Vis
spectroscopy, and COMSOL flow simulations, which together reveal how fluid dynamics and reaction
conditions govern AuNS growth mechanisms.
Our in-line UV-Vis cell provides real-time SPR spectral monitoring and provides a platform for
reinforcement learning (RL), one which we can iteratively learn and adapt the AuNS growth conditions,
such as reagent pump speeds, to achieve a target reward SPR. The RL reward function is defined as the
desired optical response and is experimentally realized through dynamic control over precursor
concentrations and Reynolds numbers via programmable pumps. This interdisciplinary effort, bridging
chemistry and computer science, lays the foundation for automated, AI-guided development of advanced
plasmonic sensing platforms.