**4. Conclusions and outlook**

In summary, the engineering of 3D multi-branched (up to 10 branches) nanostructures for sensing of analyte molecules at ultra-low concentrations, down to 1fM, is demonstrated to be highly feasible. Numerical simulations were performed to understand the underlying physics of high electric field enhancement of the plasmonic nanostructures. The advancement of the 3D fabrication methods enables the realization of uniform, homogenous and reproducible SERS devices. Reflection and SERS measurements were carried out to evaluate the MB nanostructure performances. Within this context, we demonstrated the importance of the geometry, IPS and polarization on SERS signal enhancement using 3D five-branched nanostar dimers (with sub-10 nm IPS). The elevated 3D geometry shows the advantage of high E-field enhancement over 2D geometry due to decoupling from the underlying substrate of the strong optical-near fields localized at the metal/dielectric interface. In particular, the 3D geometry enables direct interaction of analytes with hot-spot spatial regions, which are severely affected by solid dielectric substrates in the 2D geometry case. This kind of SERS architectures is particularly important in miniaturized lab-on-chip Raman detection systems, thus allowing the exploitation of lower laser powers with no consequencie over the device sensitivity. Moreover, the low-cost recycling capability of the 3D geometry counterbalances the production cost and time defined by the lithographic process. The effect of metal layer composition on SERS signal enhancement of p-MA molecules, and recycling capabilities of 3D structures were investigated with five-branched nanostar dimers in the ring structures. The effect of the number of branches, varying from 4 to 10, on the hot-spot generation and SERS enhancement was evaluated by using individual nanostructures (separated by 200 nm IPS). Moreover, the arrangement of MB nanostructures in various configurations (single, dimer, 3 × 3 array of clusters and chain of nanostructures) was evaluated to improve the device detection limit towards the single-molecule regime. 3D multi-branched nanostructures exhibit enhancement factors in the order of 1011 with an extremely high sensing capability (down to 1fM concentration). In this view, engineering the aforementioned architectures for high hot-spot density paves the way towards commercial biosensing applications, which require single/few-molecule detection sensitivity, with scalable manufacturing methods and cost-effective approaches. The proposed devices do not require specific labeling of the investigated analytes and multiple testing can be evaluated on the same platform. A new class of biological experiments will be therefore feasible, including monitoring growth factors that are produced from cultured cells. Moreover, our plasmonic nanostructures can be employed for direct detection of proteins within biological samples and real-time monitoring of chemical reactions. When applied to biomedicine, the present results, combined with already available purification methods, suggest the possibility of improving the early detection of several diseases, including cancer, where the number of clinically significant molecules at the onset of the pathology is very small and often generated by a single cell.
