Department of Chemistry, University of California, Berkeley, Berkeley, California 94720, United States.
Kavli Energy NanoSciences Institute at the University of California Berkeley and the Lawrence Berkeley National Laboratory, Berkeley, California 94720, United States.
J Phys Chem A. 2021 Feb 18;125(6):1384-1390. doi: 10.1021/acs.jpca.0c10731. Epub 2021 Feb 9.
Scanning tunneling spectroscopy (STS), a technique that records the change in the tunneling current as a function of the bias (d/d) across the gap between a tip and the sample, is a powerful tool to characterize the electronic structure of single molecules and nanomaterials. While performing STS, the structure and condition of the scanning probe microscopy (SPM) tips are critical for reliably obtaining high quality point spectra. Here, we present an automated program based on machine learning models that can identify the Au(111) Shockley surface state in d/d point spectra and perform tip conditioning on clean or sparsely covered gold surfaces with minimal user intervention. We employed a straightforward height-based segmentation algorithm to analyze STM topographic images to identify tip conditioning positions and used 1789 archived d/d spectra to train machine learning models that can ascertain the condition of the tip by evaluating the quality of the spectroscopic data. Decision tree based ensemble and boosting models and deep neural networks (DNNs) have been shown to reliably identify tips in suitable conditions for STS. We expect the automated program to reduce operational costs and time, increase reproducibility in surface science studies, and accelerate the discovery and characterization of novel nanomaterials by STM. The strategies presented in this paper can readily be adapted to STM tip conditioning on a wide variety of other common substrates.
扫描隧道谱（STS）是一种记录针尖和样品之间的间隙中偏置（d/d）函数隧穿电流变化的技术，是表征单分子和纳米材料电子结构的有力工具。在进行 STS 时，扫描探针显微镜（SPM）针尖的结构和状态对于可靠地获得高质量的点谱至关重要。在这里，我们提出了一个基于机器学习模型的自动化程序，该程序可以识别 Au(111)肖克利表面态在 d/d 点谱中的存在，并在清洁或稀疏覆盖的金表面上进行针尖处理，用户干预最小。我们采用了一种简单的基于高度的分割算法来分析 STM 形貌图像，以识别针尖处理位置，并使用 1789 个存档的 d/d 谱来训练机器学习模型，通过评估光谱数据的质量来确定针尖的状态。基于决策树的集成和提升模型以及深度神经网络（DNN）已被证明能够可靠地识别适合 STS 的条件下的针尖。我们期望自动化程序能够降低操作成本和时间，提高表面科学研究的可重复性，并通过 STM 加速新型纳米材料的发现和表征。本文提出的策略可以很容易地适应于广泛的其他常见衬底上的 STM 针尖处理。
