Scanning tunneling state recognition with multi-class neural network ensembles

Abstract

One of the largest obstacles facing scanning probe microscopy is the constant need to correct flaws in the scanning probe in situ. This is currently a manual, time-consuming process that would benefit greatly from automation. Here we introduce a convolutional neural network protocol that enables automated recognition of a variety of desirable and undesirable scanning probe tip states on both metal and non-metal surfaces. By combining the best performing models into majority voting ensembles, we find that the desirable states of H:Si(100) can be distinguished with a mean precision of 0.89 and an average receiver-operator-characteristic curve area of 0.95. More generally, high and low-quality tips can be distinguished with a mean precision of 0.96 and near perfect area-under-curve of 0.98. With trivial modifications, we also successfully automatically identify undesirable, non-surface-specific states on surfaces of Au(111) and Cu(111). In these cases we find mean precisions of 0.95 and 0.75 and area-under-curves of 0.98 and 0.94, respectively.

Embedding human heuristics in machine-learning-enabled probe microscopy

“…In addition to the relative simplicity of its reconstruction and a wealth of previous literature [25], H:Si(100) is a well understood substrate that has been used in many important advances in single atom technology and atomically precise materials engineering [24,[26][27][28][29][30][31]. Furthermore, because it has been previously studied in the context of machine-learning-enabled SPM [10][11][12], a good comparison can be formed with existing machine learning approaches based on full scans. As such, we used our existing dataset of 6167 complete images of H:Si(100) [12].…”

“…Furthermore, because it has been previously studied in the context of machine-learning-enabled SPM [10][11][12], a good comparison can be formed with existing machine learning approaches based on full scans. As such, we used our existing dataset of 6167 complete images of H:Si(100) [12]. These images were acquired on a Omicron variable-temperature STM between March 2014 and November 2015, and at varying scan sizes and voltage biases of 3×3 nm 2 to 80×80 nm 2 and −2 V to +2 V, respectively.…”

“…After random shuffling, 4987 of the images were assigned into the training/testing datasets in an 80/20 split, and 1180 into the holdout set. Data was then filtered to remove ambiguous images that were classified in multiple categories and/or the human classifiers did not perfectly agree upon [12]. This left 3386 images for training/testing, and 648 for blind verification.…”

“…Whilst convolutional neural networks (CNNs) have been shown to be capable of assessing SPM tips [10][11][12], and, most recently, of extracting 'hidden order' from STM datasets [13], CNN methods to-date have been trained exclusively with complete images. Partial scans comprising a small number of scanlines therefore simply do not provide the information upon which the network mathematically depends and so current methods of CNN image assessment require complete scans.…”

“…Beginning with a dataset of 6167 scans of the H:Si(100) surface, we extend our previous study of CNN tip detection [12] to explore and compare a variety of methods by which the state of the tip can be determined using incomplete, partial scans. In addition to the simple, common method of 'padding' incomplete scan frames with an arbitrary marker value, we also discuss training the network to recognise individual linescans instead of entire images.…”