Machine Learning Optimization of Laser-Induced Graphene Parameters for Surface-Enhanced Raman Spectroscopic Detection of Glucose

Direct laser writing on polyimide sheets offers a convenient and scalable way to produce laser-induced graphene (LIG). This is an emergent field whereby highly customizable computer-aided design provides a scalable and cost-effective method for manufacturing three-dimensional (3D) porous graphene electrodes for sensing applications. However, the difficulty in optimizing the conductivity and robustness of the graphene electrodes limits their sensing applications. Herein, we used machine-learning Bayesian optimization (BO) to optimize laser writing parameters, yielding a sheet resistance as low as ~6.75 Ω/sq. Electrochemical measurements with ferricyanide (Fe[(CN)⁶]^3-) redox molecules showed a ~45 % increase in the electroactive surface area (ESA). Furthermore, superb flexibility and minimal deformation under cyclic bending were achieved, whereby more than 98 % of the conductivity was retained after over 4,000 bending cycles, which ensures structural robustness and the long-term durability of the LIG electrodes. Silver nanoparticles (AgNPs) were deposited using voltage-controlled electrochemical deposition to add electromagnetic enhancement (EM) for surface-enhanced Raman spectroscopy (SERS) measurements. The AgNPs deposited LIGs were functionalized with glucose-binding 4-Mercaptophenylboronic acid (4-MPBA) molecules, utilizing strong Ag-thiol covalent interaction, and formed self-assembled monolayers (SAM) on the LIG surface, effectively shielding non-specific adsorptions of the glucose molecules on the electrode surface. The 4-MPBA-AgNPs@LIG glucose sensors showed a linear response and high selectivity to the glucose (0-5 mM) at 7.5 pH. Our results provide an efficient way to rapidly optimize the laser parameters and tailor the surface chemistry for designing miniaturized flexible LIG sensors for selective glucose detection.