Leveraging Artificial Intelligence and Reverse Tip Sample Configuration for Automation of Data Processing in Quantitative Scanning Spreading Resistance Microscopy

Abstract

Scanning probe microscopy (SPM) has become a vital metrology tool for characterizing nanoscale devices with exceptional spatial resolution, driving advances in various fields. However, its low overall throughput remains a major limitation. To address this, the high-efficiency data acquisition capabilities of reverse tip sample (RTS) SPM are combined with automated data processing via artificial intelligence (AI)-based computer vision algorithms. The effectiveness of this approach is demonstrated through a case study of scanning spreading resistance microscopy (SSRM). YOLO (You Only Look Once) models are trained to detect each layer in SSRM resistance maps of a calibration sample, serving as a key step in automating the quantitative SSRM data processing workflow. Models trained on mixed datasets of standard and RTS SPM images (ratio 1:4) achieve an excellent accuracy of 97.8%, while reducing the data collection time fivefold compared to using solely standard calibration datasets. Additionally, the model's strong ability to effectively recognize and exclude measurement artifacts during layer selection further enhances its suitability for real-world applications. This work significantly accelerates the SSRM data analysis by automating the workflow and highlights the potential of RTS SPM as a high-throughput solution for generating AI training data, facilitating faster AI model deployment in SPM applications.