[Paper Review] Autonomous Scanning Probe Microscopy in-situ Tip Conditioning through Machine Learning

Mohammad Rashidi, Robert A. Wolkow|PubMed|Mar 19, 2018

Force Microscopy Techniques and Applications7 citations

TL;DR

This paper presents a machine learning-based system that autonomously detects and conditions scanning probe microscope (SPM) tips in situ using a convolutional neural network (CNN) to assess tip quality via imaging silicon dangling bonds. The method achieves 99%+ accuracy in identifying degraded tips and automatically performs in-situ conditioning until a sharp tip is restored, enabling reliable atomic-scale fabrication without manual intervention.

ABSTRACT

Atomic-scale characterization and manipulation with scanning probe microscopy rely upon the use of an atomically sharp probe. Here we present automated methods based on machine learning to automatically detect and recondition the quality of the probe of a scanning tunneling microscope. As a model system, we employ these techniques on the technologically relevant hydrogen-terminated silicon surface, training the network to recognize abnormalities in the appearance of surface dangling bonds. Of the machine learning methods tested, a convolutional neural network yielded the greatest accuracy, achieving a positive identification of degraded tips in 97% of the test cases. By using multiple points of comparison and majority voting, the accuracy of the method is improved beyond 99%.

Motivation & Objective

• To automate in-situ tip conditioning in scanning probe microscopy to eliminate manual intervention and improve reliability in atomic-scale experiments.
• To develop a machine learning framework that can detect tip degradation by analyzing STM images of surface dangling bonds.
• To integrate automated tip quality assessment and conditioning into existing autonomous atomic fabrication workflows.
• To generalize the method to other material systems and nanoscale imaging techniques.

Proposed method

• A convolutional neural network (CNN) is trained on 3,500 manually labeled 28×28 pixel STM images of isolated silicon dangling bonds on a hydrogen-terminated Si(100) surface.
• The training dataset is augmented by rotating and mirroring each image, increasing the dataset size by eightfold to improve model generalization.
• The CNN evaluates tip quality by analyzing the appearance of isolated dangling bonds in real-time STM images during operation.
• When the CNN detects a double tip (degraded quality), the system triggers in-situ tip conditioning via controlled indentation at a preselected surface spot.
• Multiple image comparisons and majority voting across several dangling bond images are used to enhance classification accuracy beyond 99%.
• The automation routine is implemented in Python and LabVIEW using the Nanonis SPM controller interface, integrated into a full SPM fabrication pipeline.

Experimental results

Research questions

• RQ1 Can a machine learning model accurately detect degraded scanning probe microscope tips based on STM images of surface defects?
• RQ2 How does majority voting across multiple image samples improve the reliability of tip quality classification?
• RQ3 Can automated in-situ tip conditioning restore atomic sharpness without human supervision?
• RQ4 To what extent can this framework be generalized to other materials and nanoscale imaging techniques?

Key findings

• The convolutional neural network achieved a precision score of 97% in identifying degraded tips from STM images of silicon dangling bonds.
• By applying majority voting across multiple images, the system improved tip quality classification accuracy to over 99%.
• The automated routine successfully detected a double tip after atomic fabrication and restored tip sharpness after three conditioning cycles, as confirmed by subsequent STM imaging.
• The method was successfully integrated into an autonomous atomic wire fabrication process, maintaining high image quality and low error rates throughout the patterning sequence.
• The framework demonstrated robust performance on a real-world SPM system operating at 4.5 K under ultrahigh vacuum, using electrochemically etched tungsten tips.
• The approach is generalizable to other material systems and nanoscale imaging techniques, including critical dimension analysis in semiconductor manufacturing.