Prediction and assessment of nanoprobe for tip-enhanced Raman spectroscopy: Data-driven artificial intelligence approach

Abstract

In Tip-enhanced Raman spectroscopy (TERS), the geometry of the metallic tip critically governs the strength of localized surface plasmon resonance and the achievable spatial resolution, yet reliable pre-experimental quality assessment remains challenging. Conventional electrochemical etching of gold nanoprobes produces substantial geometric variability, forcing researchers to rely on post-fabrication scanning electron microscopy (SEM) verification, a costly and labor-intensive bottleneck that limits high-throughput TERS experiments. Here, we introduce an integrated explainable artificial intelligence (XAI) framework that predicts tip geometry directly from real-time electrochemical etching current signals, achieving a mean absolute percentage error of 9.47 %. Through XAI perturbation analysis, we identify the final segment of the etching current trajectory as the most informative region for predicting nanoscale tip geometry. The predicted geometries are further evaluated through finite-difference time-domain (FDTD) simulations, and experimental validation is performed using scanning tunneling microscopy-based TERS measurements on monolayer WS2. By enabling rapid screening of candidate probes, the proposed framework reduces the reliance on routine post-fabrication SEM characterization and shortens the nanoprobe development cycle, thereby improving experimental efficiency and enhancing the scalability of TERS probe development.