AI method breaks the barrier, understanding microscopic images

Atomic force microscopy, or AFM, is a widely used technique that can measure material surfaces in three dimensions, but its accuracy is limited by the microscope's probe size. A new AI technique overcomes this limitation, allowing microscopes to resolve matter

Features smaller than probe tip.

A deep learning algorithm developed by researchers at the University of Illinois Urbana-Champaign was trained to remove the effects of probe width from AFM microscope images. As reported in the journal Nano Letters, the algorithm outperforms other methods in providing the first truly three-dimensional surface profiles at resolutions below the width of a microscope probe tip.

„Accurate surface height profiles are important for the development of nanoelectronics and scientific studies of material and biological systems, and AFM is a key technique that can measure the profiles non-invasively,” he said. Yingzhi ZhangA U. Materials Science and Engineering Professor and Project Leader. „We've demonstrated how to be more precise and see even smaller things, and we've shown how AI can be used to overcome a seemingly insurmountable limit.”

Often, microscopic techniques can only provide two-dimensional images, essentially providing researchers with aerial photographs of object surfaces. AFM provides full topographic maps that accurately show the height profiles of surface features. These three-dimensional images are obtained by moving a probe over the object's surface and measuring its vertical deflection.

If the surface features approach the size of the tip of the probe – about 10 nanometers – then they cannot be resolved by the microscope because the probe becomes too large to „feel” the features. Microscopists have known about this limitation for decades, but I. Researchers at the U.

READ  New Hubble telescope image shows mysterious spokes in Saturn's rings

„We turned to AI and deep learning because we wanted to get the height — the precision rigor — without the inherent limitations of more conventional mathematical methods,” said Lalit Bonagiri, a graduate student in Zhang's group and lead author of the study.

Researchers have developed a deep learning algorithm with an encoder-decoder architecture. It first “codes” raw AFM images by decomposing them into abstract features. After the feature representation has been manipulated to remove undesirable effects, it is again „decoded”.

A recognizable image.

To train the algorithm, the researchers created synthetic images of three-dimensional structures and simulated their AFM readings. An algorithm was constructed to transform simulated AFM images with probe size effects and extract fundamental features.

„We had to do something really buggy to achieve this,” Bonagiri said. „The first step in typical AI image processing is to rescale the brightness and contrast of images against some standard to simplify comparisons. In our case, the absolute brightness and contrast were the meaningful part, so we had to drop that first. This made the problem more challenging.”

To test their methodology, the researchers synthesized gold and palladium nanoparticles of known dimensions on a silicon host. The algorithm successfully removed the probe tip effects and correctly identified the three-dimensional features of the nanoparticles.

„We've provided a proof of concept and shown how to use AI to significantly improve AFM imaging, but this work is just the beginning,” Zhang said. „As with all AI algorithms, it can be improved by training on more and better data, but the path forward is clear.”

Dodaj komentarz

Twój adres e-mail nie zostanie opublikowany. Wymagane pola są oznaczone *