Abstract
The rapid development of methods for scanning magnetic metrology at the nanoscale has greatly improved our ability to study the electric properties of low-dimensional materials and interfaces, including superconducting thin films and electronic devices. However, existing methods limit the imaging of electric current from magnetic images. This study provides a computational method that lifts these limitations, providing a more accurate solution of the inversion problem. This approach has the potential to enhance magnetic imaging of electrical transport in a variety of nanoscale two-dimensional systems.
Introduction
Mapping the flow of electricity through thin films—such as superconductors, topological insulators, and graphene—is essential for the development of next-generation electronic devices. Researchers typically achieve this by measuring the magnetic field ($B_z$) just above a sample and "inverting" the Biot-Savart law to determine the underlying current density ($J$).
However, this inversion is a mathematically "ill-posed" problem. Because the magnetic field decays rapidly with distance, small amounts of measurement noise are exponentially amplified during the reconstruction process. Traditional approaches attempt to mitigate this by using a two-step "Global Inversion" (GI) process: first calculating a scalar stream function ($g$) and then differentiating that function to find the current. This methodology often requires filtering at both stages, which "oversmoothes" the resulting data and washes out sharp features like narrow current filaments. Furthermore, standard algorithms struggle with "transport currents"—currents that flow in from one side of the measurement window and out the other—often producing significant artifacts at the boundaries.
The Direct Inversion (DI) Framework
To address these limitations, this research introduces a Direct Inversion (DI) scheme. Rather than relying on an intermediate stream function, the DI method solves for the $x$ and $y$ components of the current density simultaneously in a single computational step. This approach offers several technical advantages:
- Single-Step Regularization: By skipping the stream function, the DI method applies only one regularization filter. This preserves sharp edges and fine spatial details that traditional methods blur away.
- Tikhonov Regularization: To manage noise, the researchers utilized Tikhonov regularization, a damping technique that penalizes non-smooth solutions.
- Automated Parameter Selection: The framework implements two robust methods for automatically determining the optimal level of smoothness: Generalized Cross-Validation (GCV) and Stein’s Unbiased Risk Estimate (SURE).
- Analytical Kernels: The method utilizes exact analytical expressions for magnetic kernels rather than discrete approximations. This is particularly critical for high-resolution imaging where the sensor is positioned very close to the sample surface.
Overcoming Boundary Artifacts: The Reflection Rule
A major hurdle in magnetic imaging is the assumption that current must form a closed loop within the camera's field of view. When studying real-world devices, current often crosses the measurement boundary. To solve this, the authors developed a Reflection Rule. By symmetrically mirroring the magnetic field image, the algorithm "tricks" the Discrete Fourier Transform (DFT) into treating the field as continuous and periodic. This eliminates the "Gibbs oscillations" (ripples) that typically ruin data at the edges of the measurement window, allowing for the accurate reconstruction of transport currents.
Experimental Validation
The DI method was benchmarked using numerically simulated datasets to compare reconstructions against a known "ground truth" across three distinct scenarios:
- Sample A: A simple closed square loop.
- Sample B: A complex loop where current flows in opposite directions in the inner and outer regions, with the measurement window capturing only a corner.
- Sample C: A loop containing superconducting vortices, where the measurement window cuts through the loop on all four sides.
Tests were conducted across multiple noise levels. In low-noise scenarios, the DI method achieved a Mean Square Deviation (MSD) more than an order of magnitude lower than traditional stream-function methods. Notably, the DI-SURE approach proved particularly robust when the sensor was further from the sample or when dealing with irregular transport currents.
Robustness and Resolution
Beyond accuracy, the DI method demonstrated significant resilience to experimental uncertainties, such as errors in estimating the probe's height ($h$). Even with slight underestimations of height, the reconstruction remained remarkably stable. This robustness is a significant benefit for experimentalists working at the nanometer scale, where precise positioning is a constant challenge. Ultimately, the DI method provides a "sharper lens," making it superior for identifying one-dimensional transport paths at material interfaces.
Conclusion and Future Outlook
The Direct Inversion method represents a significant step forward in nanoscale magnetometry, offering higher spatial resolution and the ability to handle complex boundary conditions. While the current model is optimized for two-dimensional current distributions in thin films, it provides a foundation for future work. Potential expansions include adapting the math for 3D volume currents and exploring high-frequency AC transport. By providing a more accurate and user-friendly inversion tool, this method promises to enhance our understanding of electrical transport in a wide array of low-dimensional systems.
Summary updated