Feb 29 2024Reviewed by Lexie Corner
The accuracy of atomic force microscopy (AFM), a widely used technique that maps material surfaces quantitatively in three dimensions, is limited by the size of the probe on the microscope.
This restriction is removed by a novel AI method, which enables microscopes to resolve material features smaller than the tip of the probe.
Researchers at the University of Illinois Urbana-Champaign trained a deep learning algorithm to eliminate probe width effects from AFM microscope images. According to research published in the journal Nano Letters, the algorithm provides the first accurate three-dimensional surface profiles at resolutions smaller than the width of the microscope probe tip, outperforming alternative approaches.
Accurate surface height profiles are crucial to nanoelectronics development as well as scientific studies of material and biological systems, and AFM is a key technique that can measure profiles noninvasively. We’ve demonstrated how to be even more precise and see things that are even smaller, and we’ve shown how AI can be leveraged to overcome a seemingly insurmountable limitation.
Yingjie Zhang, Professor and Project Lead, Materials Science and Engineering, University of Illinois
Frequently, microscopy techniques are limited to providing two-dimensional images, essentially offering researchers aerial photographs of material surfaces. AFM provides complete topographical maps with precise height profiles of the surface features. By advancing a probe across the material's surface and measuring its vertical deflection, these three-dimensional images are produced.
A microscope cannot resolve surface features if they get close to the probe's tip size, which is around 10 nm, as the probe becomes too big to "feel out" the features. Although this limitation has been known to microscopists for decades, the researchers from the University of Illinois are the first to provide a deterministic solution.
We turned to AI and deep learning because we wanted to get the height profile – the exact roughness – without the inherent limitations of more conventional mathematical methods.
Lalith Krishna Samanth Bonagiri, Graduate Student and Study Lead Author, University of Illinois
The researchers created an encoder-decoder framework and a deep learning algorithm. By breaking down raw AFM images into abstract features, it first "encodes" the images. Subsequently, an algorithm was developed to convert the simulated AFM images, accounting for probe-size effects, and to extract the underlying features.
The scientists created synthetic images of three-dimensional structures and mimicked their AFM readouts to train the algorithm. Next, an algorithm was built to extract the underlying features and transform the simulated AFM images with probe-size effects.
We actually had to do something nonstandard to achieve this. The first step of typical AI image processing is to rescale the brightness and contrast of the images against some standard to simplify comparisons. In our case, though, the absolute brightness and contrast is the part that’s meaningful, so we had to forgo that first step. That made the problem much more challenging.
Lalith Krishna Samanth Bonagiri, Graduate Student and Study Lead Author, University of Illinois
To test their algorithm, the researchers created known-dimension gold and palladium nanoparticles on a silicon host. The algorithm correctly identified the three-dimensional features of the nanoparticles and successfully eliminated the effects of the probe tip.
Zhang concludes, “We’ve given a proof-of-concept and shown how to use AI to significantly improve AFM images, but this work is only the beginning. As with all AI algorithms, we can improve it by training it on more and better data, but the path forward is clear.”
Journal Reference:
Bonagiri, S. K. L., et al. (2024) Precise Surface Profiling at the Nanoscale Enabled by Deep Learning. Nano Letters. doi.org/10.1021/acs.nanolett.3c04712
Source: https://grainger.illinois.edu/