Deep learning for enhancement of low-resolution and noisy scanning probe microscopy images

• Samuel Gelman,
• Irit Rosenhek-Goldian,
• Nir Kampf,
• Marek Patočka,
• Maricarmen Rios,
• Marcos Penedo,
• Georg Fantner,
• Amir Beker,
• Sidney R. Cohen and
• Ido Azuri

Beilstein J. Nanotechnol. 2025, 16, 1129–1140, doi:10.3762/bjnano.16.83

• shows deep learning models to be superior for super-resolution tasks and enables significant reduction in AFM measurement time, whereby low-pixel-resolution AFM images are enhanced in both resolution and fidelity through deep learning. Keywords: atomic force microscopy; deep learning; fast scanning

• interactions. Scanning distortions due to non-linearities in the scan are also trickier to correct and harder to avoid in fast scanning. Some of these artifacts can be eliminated or attenuated by image processing techniques [4][5][6][7][8][9][10][11][12]. Another resolution-limiting factor in AFM is the tip

• deep learning models in comparison to the traditional methods. This could arise from the difference between fast and traditional AFM scanning as Celgard® 2400 was measured with fast AFM and titanium film with standard, slow AFM. The fast-scanning is often achieved at the expense of somewhat worse

Shape, membrane morphology, and morphodynamic response of metabolically active human mitochondria revealed by scanning ion conductance microscopy

• Eric Lieberwirth,
• Anja Schaeper,
• Regina Lange,
• Ingo Barke,
• Simone Baltrusch and
• Sylvia Speller

Beilstein J. Nanotechnol. 2025, 16, 951–967, doi:10.3762/bjnano.16.73

• . Difference images were colour-coded with a custom colour pattern, where the centre of the scale is white, positive heights are red, and negative heights are blue. Line profiles were always extracted in Gwyddion along the fast scanning direction (x-axis). The line profile data was read and processed in the

The role of convolutional neural networks in scanning probe microscopy: a review

• Ido Azuri,
• Irit Rosenhek-Goldian,
• Neta Regev-Rudzki,
• Georg Fantner and
• Sidney R. Cohen

Beilstein J. Nanotechnol. 2021, 12, 878–901, doi:10.3762/bjnano.12.66

• molecules, the test set can be generated computationally from first principles. Various fast-scanning approaches have also been applied. Augmentation techniques are often required. Another characteristic of SPM is that it is multimodal. One scan can provide multiple mappings simultaneously with the

Reducing molecular simulation time for AFM images based on super-resolution methods

• Zhipeng Dou,
• Jianqiang Qian,
• Yingzi Li,
• Rui Lin,
• Jianhai Wang,
• Peng Cheng and
• Zeyu Xu

Beilstein J. Nanotechnol. 2021, 12, 775–785, doi:10.3762/bjnano.12.61

• the molecular structure [26], recognizing and classifying surface features [27][28][29], and fast scanning [30][31]. The main problem in training models of machine learning is providing sufficiently labeled training data. High-resolution AFM experiments are time consuming and experimental data are, a

Highly ordered mesoporous silica film nanocomposites containing gold nanoparticles for the catalytic reduction of 4-nitrophenol

• Mohamad Azani Jalani,
• Leny Yuliati,
• Siew Ling Lee and
• Hendrik O. Lintang

Beilstein J. Nanotechnol. 2019, 10, 1368–1379, doi:10.3762/bjnano.10.135

• (ICP-OES) on an Agilent model 700 series spectrometer where the samples were digested in 5 wt % aqua regia. The catalytic performance was studied via a Shimadzu DR UV–vis spectrophotometer (UV-2600) under liquid set-up with the fast scanning method using a standard 3 mL quartz cell. Fabrication of

Pure and mixed ordered monolayers of tetracyano-2,6-naphthoquinodimethane and hexathiapentacene on the Ag(100) surface

• Robert Harbers,
• Timo Heepenstrick,
• Dmitrii F. Perepichka and
• Moritz Sokolowski

Beilstein J. Nanotechnol. 2019, 10, 1188–1199, doi:10.3762/bjnano.10.118

• hydrogen bonds (green lines). (c) Drift-corrected and filtered STM image. The fast scanning direction is the vertical direction (Ub = −0.81 V; It = 127 pA). A hard-sphere model of the structure derived from LEED has been overlaid. The non-symmetric shape of the molecule is assigned to tip effects, but not

• showing different islands. Here, all four possible domains with different orientations can be observed at the same time. They are marked by A–D. The fast scanning direction is in horizontal orientation (Ub = −1.51 V; It = 82 pA). This STM image has been selected to demonstrate the presence of the

• different symmetry-equivalent domains. The corresonding islands are smaller than usual. (c) STM image. The fast scanning direction is horizontal (Ub = −0.80 V; It = 80 pA). A hard-sphere model of the structure derived from LEED has been superimposed. (d) Hard-sphere model with indicated unit cell and

Biomimetic surface structures in steel fabricated with femtosecond laser pulses: influence of laser rescanning on morphology and wettability

• Camilo Florian Baron,
• Alexandros Mimidis,
• Daniel Puerto,
• Evangelos Skoulas,
• Emmanuel Stratakis,
• Javier Solis and
• Jan Siegel

Beilstein J. Nanotechnol. 2018, 9, 2802–2812, doi:10.3762/bjnano.9.262

• right combination of a high repetition rate laser and fast scanning head. The specific experimental conditions used to fabricate the different areas are displayed in Table 1, including 4 different laser fluence values. SEM micrographs of the 16 different areas corresponding to the experimental

Optical near-field mapping of plasmonic nanostructures prepared by nanosphere lithography

• Gitanjali Kolhatkar,
• Alexandre Merlen,
• Jiawei Zhang,
• Chahinez Dab,
• Gregory Q. Wallace,
• François Lagugné-Labarthet and
• Andreas Ruediger

Beilstein J. Nanotechnol. 2018, 9, 1536–1543, doi:10.3762/bjnano.9.144

• tips to ensure a detectable near-field contribution that will exceed the ubiquitous background far-field signal in the scattered light. In ideal conditions, fast scanning on the surface ensures a high sensitivity, and minimizes the drift. To achieve this objective, most nano-antennas used in aSNOM are

Automated image segmentation-assisted flattening of atomic force microscopy images

• Yuliang Wang,
• Tongda Lu,
• Xiaolai Li and
• Huimin Wang

Beilstein J. Nanotechnol. 2018, 9, 975–985, doi:10.3762/bjnano.9.91

• flattening During scanning, AFM images are constructed line by line. The time required for each scan line is around 1 s along the fast scanning direction, which is much shorter than that of several minutes required for an entire image. As a result, the direction of drift is normally perpendicular to that of

• the fast scanning, namely, along the slow scan direction. Along individual scan lines, the drift is much lower. Here we investigate the influence of fitting direction on the performance of image flattening in SWCF. Figure 8a shows a raw AFM image. The SWCF was then applied to individual scan lines

• along the horizontal direction (fast scanning direction). The flattened result is shown in Figure 8b. One can see that the SWCF along the horizontal direction gives an optimized image flattening. To test the influence of fitting direction relative to the drift direction, here the raw AFM image was

Direct writing of gold nanostructures with an electron beam: On the way to pure nanostructures by combining optimized deposition with oxygen-plasma treatment

• Domagoj Belić,
• Mostafa M. Shawrav,
• Emmerich Bertagnolli and
• Heinz D. Wanzenboeck

Beilstein J. Nanotechnol. 2017, 8, 2530–2543, doi:10.3762/bjnano.8.253

• atmosphere for a few hours during the transfer from the fabrication facility to the TEM lab. AFM investigations were carried out on a Veeco/Bruker Dimension 3000 atomic force microscope. (a) Schematics of the FEBID experimental set-up. Note that the direction of electron beam movement, i.e., fast scanning

Nanoscale rippling on polymer surfaces induced by AFM manipulation

• Mario D’Acunto,
• Franco Dinelli and
• Pasqualantonio Pingue

Beilstein J. Nanotechnol. 2015, 6, 2278–2289, doi:10.3762/bjnano.6.234

• pattern. In this case the observed ripples were perpendicular or almost perpendicular to the fast scanning direction. The researchers have soon realized that many parameters can affect the formation of nanoripples. These parameters depend on the characteristics of the tip–surface contact, the experimental

Kelvin probe force microscopy for local characterisation of active nanoelectronic devices

• Tino Wagner,
• Hannes Beyer,
• Patrick Reissner,
• Philipp Mensch,
• Heike Riel,
• Bernd Gotsmann and
• Andreas Stemmer

Beilstein J. Nanotechnol. 2015, 6, 2193–2206, doi:10.3762/bjnano.6.225

• ] or fast scanning [39]. According to the separation principle [37], the optimal controller that minimises the expected error can be constructed by finding an optimal ‘observer’ and an optimal ‘regulator’. As an observer, we use a Kalman filter [40], which continuously blends the sideband measurements

Scanning reflection ion microscopy in a helium ion microscope

• Yuri V. Petrov and
• Oleg F. Vyvenko

Beilstein J. Nanotechnol. 2015, 6, 1125–1137, doi:10.3762/bjnano.6.114

• advantage as SEM: fast scanning and a large maximum field of view. On the other hand, scanning and measurements in AFM can be performed for all sides of the sample features, whereas in RIM (as well as in REM) only the side facing the beam is suited for investigation. In this work, RIM was demonstrated to