Trace-level detection and classifications of pentaerythritol tetranitrate via geometrically optimized film-based Au/ZnO SERS sensors

This work has explored the scalable, multi-batch fabrication process of the Au-decorated ZnO nanorod (NR) SERS sensors with five designed geometries (A to E), based on the hydrothermally synthesized ZnO NR templates prepared at different reaction time (3.0, 4.5, 6.0, 8.0, and 10.0 h). Each sample geometry was fabricated for three batches. The physical morphologies of the templates and the SERS substrates were examined by various techniques, whose results demonstrated the variations in their physical dimensions. The aggregations of the nanorods into several clusters at the tips were clearly observed. The material compositions were verified with ZnO as the inner axial rod, and the Au element enveloping the nanorod tip. The finite element method (FEM) simulations predicted the generated E-field at 3.32 V/m obtained from the sample geometry B (4.5 h). The SERS performance were thoroughly investigated with the pentaerythritol tetranitrate (PETN) substance obtained from three different sources, each with serial dilution from 100 mg/ml to 0.1 µg/ml. The analyte solution was drop-casted on the SERS sensors, on which the hyperspectral Raman-mapping measurements were performed. With principal component analysis (PCA) and linear discriminant analysis (LDA), the results showed that the most optimal SERS sensors yielded distinguishable clusters of the different PETN sources, with the classification accuracy at 73.47 ± 0.64%. The proposed SERS sensors exhibited the optimal performance at 10−4 g/ml, the limit of detection (LOD) at 10−7 g/ml with the enhancement factor (EF) of 3.01 × 106, which therefore allowed potential implementation towards the trace detection and classifications of the PETN substance. •Au/ZnO SERS sensors of 5 designed geometries were fabricated (multi-batch).•PETN analytes obtained during search and seizure from 3 sources were tested.•Optimal designs yielded LOD at 0.10 ppm and EF at 3.01×106.•Single chip yielded the smallest RSD at 2.43%, and multi chips at 8.23%.•Successful classifications were obtained by machine learning with 74% accuracy.