Multitask Scanning Probe Microscopy

Scanning probe microscopy (SPM) is essential for characterizing materials at the nanoscale, but applying it to large-scale samples like semiconductor wafers or combinatorial libraries presents a significant challenge. Exhaustive mapping of multiple properties across thousands of locations is often impractical due to time constraints and the risk of sample damage. This paper addresses the problem of how to efficiently allocate a limited measurement budget when choosing between different experimental protocols (e.g., tapping-mode vs. contact-mode) and spatial locations.

The researchers introduce "multitask scanning probe microscopy," a closed-loop, autonomous workflow. Instead of treating each SPM modality as an independent experiment, the system models them as related tasks within a multitask Gaussian process (GP). By using an intrinsic coregionalization model, the system learns the spatial structure of the material and the statistical correlations between different measurement modes.

At each step, the controller predicts the response and uncertainty for all tasks across the entire sample. It then uses a random-scalarization upper-confidence-bound (UCB) acquisition function to select the next location and the most informative measurement protocol. This allows the microscope to "learn" the relationship between, for example, tapping-mode topography and contact-resonance measurements, enabling it to update the response landscape for both modes even when only one is performed at a specific site.

The framework was implemented on an automated atomic force microscope and tested on a composition-spread AlScN wafer. The system successfully navigated the wafer, autonomously switching between tapping-mode and Dual AC Resonance Tracking (DART) protocols. The results demonstrate that the multitask model effectively transfers information between modalities, allowing for a more efficient reconstruction of surface property maps compared to traditional, fixed-protocol scanning. The approach provides a scalable foundation for combining rapid, non-destructive imaging with more complex, time-consuming spectroscopic or electrical measurements.