Oak Ridge National Laboratory researchers have developed SimuScan, an artificial intelligence framework that guides atomic force microscopes toward the most informative areas of a sample for closer study. Although atomic force microscopy reveals structures as small as molecules, operating the instrument traditionally requires expert judgment regarding scan location and settings; SimuScan aims to reduce this burden and enable faster, more consistent research.
“Operating an atomic force microscope is a bit like piloting a modern jet,” said Liam Collins, an ORNL senior R&D scientist at the Center for Nanophase Materials Sciences (CNMS). “The hardware has incredible capability, but making full use of it often requires an experienced pilot.” The system addresses a key challenge in AI-assisted microscopy by training on synthetic data, bypassing the need for large datasets of real images numbering in the thousands.
SimuScan Leverages Synthetic Data to Address AFM Labeling Challenges
SimuScan addresses a critical limitation in atomic force microscopy by generating realistic synthetic images that include common imperfections such as tip effects, scanner drift, and surface roughness, all recreated within the simulation. This approach bypasses the need for extensive, manually labeled datasets of real AFM images, a process that can consume weeks of expert time and introduce variability between different researchers. The system can produce thousands of labeled images with controlled variations in object shapes and backgrounds, allowing experimental data to be used primarily for model testing and refinement.
Researchers note that the challenge extends beyond simply acquiring an image to understanding its contents and prioritizing where the microscope should scan next. They explain that the difficulty lies not only in acquiring the image but also in understanding what is in the image, deciding what matters, and knowing where the microscope should look next, highlighting the cognitive load currently placed on AFM operators.
SimuScan aims to automate this decision-making process, directing the instrument toward regions most likely to contain scientifically relevant information. Validating the system involved training AI models on synthetic images and then assessing their ability to accurately identify features in real AFM data obtained from fabricated nanostructures, DNA assemblies, and bacterial cells.
Realism in the synthetic data is paramount; the team likened the process to flight simulation. “The true test of realism is whether an AI trained in a flight simulator can successfully land a real plane in a storm,” said Liam Collins, emphasizing the need for the AI to generalize from simulated to actual laboratory conditions.
The researchers discovered that the largest sources of error often originate in the background of the images, rather than the target features themselves, a finding that informs ongoing refinements to the simulation. They noted that tip geometry, drift, flattening, and contamination can all introduce artifacts that resemble real nanoscale structures, and experienced users learn to distinguish them; AI models must be taught to do the same, underscoring the complexity of accurately interpreting AFM data.
Operating an atomic force microscope is a bit like piloting a modern jet.
Liam Collins, an ORNL senior R&D scientist at the Center for Nanophase Materials Sciences (CNMS)
AI-Guided AFM Targets Informative Nanoscale Features Autonomously
This approach allows the AI to autonomously identify and prioritize nanoscale features during analysis, a capability previously reliant on expert interpretation of complex scans. The framework supports a closed-loop imaging process, beginning with a broad, low-resolution scan of a sample’s surface. SimuScan then segments identified features, ranks them according to criteria set by the user, and directs the microscope to areas most likely to yield scientifically valuable data.
This targeted approach contrasts with conventional AFM, where researchers manually select scan locations and settings based on visual inspection of initial images. “A camera records reflected light, but an atomic force microscope acts more like a high-tech record player needle feeling its way across a landscape,” explained Liam Collins. This analogy highlights that AFM data is not a direct representation of the sample, but a product of the interaction between the probe and the surface.
A camera records reflected light, but an atomic force microscope acts more like a high-tech record player needle feeling its way across a landscape.
Liam Collins, an ORNL senior R&D scientist at the Center for Nanophase Materials Sciences (CNMS)
Transfer Learning Validates SimuScan Across Real Biological Samples
Atomic force microscopy data, unlike photographs, inherently blends sample characteristics with the specifics of the measurement itself. The team’s validation process extended beyond fabricated materials to include biological samples like DNA assemblies and bacterial cells, demonstrating the framework’s adaptability. This focus on realistic imperfections allows AI models trained on synthetic data to accurately identify features within actual AFM data. SimuScan’s closed-loop imaging process further refines data acquisition by prioritizing regions of interest.
Beginning with a broad, low-resolution scan, the AI segments potential features, ranks them according to user-defined criteria, and then directs the microscope for closer examination. This targeted approach shifts the researcher’s role from manually searching for features to interpreting the results, increasing efficiency and allowing for more complex analyses.
The challenge is not only acquiring the image but also understanding what is in the image, deciding what matters and knowing where the microscope should look next.
Ruben Millan Solsona, an ORNL technical professional and staff scientist
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
