Performance improvement of single-molecule sensors through deep learning-based decoding of tunneling signals enables sub-attomolar sensitivity.

Abstract

Detecting individual molecules in real time provides high sensitivity for sensing applications. The break junction technique enables highly sensitive single-molecule detection by capturing the specific electronic signatures of individual molecules. However, this method is typically restricted by the requirement for anchoring groups on target analytes. Harnessing intermolecular interactions offers a solution to detect molecules without anchoring groups. Yet the resulting signals are often weak, sparse, and hidden in ensemble analyses. Here, we integrate rationally designed porphyrin-based probes with a time-frequency deep-learning framework to decode these subtle signatures. With this approach, we elevate the detection limit to the sub-attomolar level (10^-18 mol L^-1) with seconds-scale response time (26 s). Validated across diverse analytes, our strategy consistently enhances sensitivity, confirming its generalizability. This synergistic strategy establishes a paradigm for single-molecule chemical sensing both in methodological and performance dimensions. By transforming fleeting interactions into actionable detection, the approach could provide a robust and broadly applicable framework for environmental monitoring and molecular diagnostic.