SparseC-AFM: a deep learning method for fast and accurate characterization of MoS$_2$ with C-AFM

TL;DR

The paper addresses the bottleneck of slow C-AFM metrology for 2D materials like $MoS_2$ by developing SparseC-AFM, an AI-accelerated workflow that reconstructs full-resolution morphology and current maps from undersampled scans. The method uses a SwinIR-based upsampling network to achieve up to $11\times$ faster data acquisition with preserved electrical-property predictions, demonstrated across sparsity factors up to $\times64$. It outperforms prior sparse AFM approaches in reconstruction quality (PSNR/SSIM) while maintaining key metrics such as film coverage and defect density, enabling reliable material characterization. The approach is non-intrusive and generalizable, offering a practical path for industrial 2D-material metrology, with code and weights available at GitHub.

Abstract

The increasing use of two-dimensional (2D) materials in nanoelectronics demands robust metrology techniques for electrical characterization, especially for large-scale production. While atomic force microscopy (AFM) techniques like conductive AFM (C-AFM) offer high accuracy, they suffer from slow data acquisition speeds due to the raster scanning process. To address this, we introduce SparseC-AFM, a deep learning model that rapidly and accurately reconstructs conductivity maps of 2D materials like MoS$_2$ from sparse C-AFM scans. Our approach is robust across various scanning modes, substrates, and experimental conditions. We report a comparison between (a) classic flow implementation, where a high pixel density C-AFM image (e.g., 15 minutes to collect) is manually parsed to extract relevant material parameters, and (b) our SparseC-AFM method, which achieves the same operation using data that requires substantially less acquisition time (e.g., under 5 minutes). SparseC-AFM enables efficient extraction of critical material parameters in MoS$_2$, including film coverage, defect density, and identification of crystalline island boundaries, edges, and cracks. We achieve over 11x reduction in acquisition time compared to manual extraction from a full-resolution C-AFM image. Moreover, we demonstrate that our model-predicted samples exhibit remarkably similar electrical properties to full-resolution data gathered using classic-flow scanning. This work represents a significant step toward translating AI-assisted 2D material characterization from laboratory research to industrial fabrication. Code and model weights are available at github.com/UNITES-Lab/sparse-cafm.

Figures (6)

• Figure 1: Classic flow implementation for full-resolution mapping of MoS$_2$ with C-AFM. a) Conductive AFM tip sweeps an MoS$_2$ surface as an amplifier (A) records current response under an applied bias. b) Path traced by a conductive tip; each dot represents a sampling point on a dense grid. c) Cross-section of the junction between conductive tip and material surface. $\sim$10 nm-radius tip (grey circles) makes contact with MoS$_2$ lattice (yellow circles) with normal loading force $F$ and current path $e^{-}$. d) Full-resolution dataset produced by classic-flow implementation. (left) (256-512)$\times$(256-512)-pixel current map of monolayer MoS$_2$ visualized in RGB color-space. (Center) corresponding binary pixel map showing every probe location; (right, inset) magnified unit cell.
• Figure 2: Our proposed workflow for rapid C-AFM scanning and characterization of 2D materials with SparseC-AFM. (Top-left) a high-level figure of our data acquisition pipeline for C-AFM. (Bottom-left) standard, full-resolution data collection and analysis procedure. (Right) our proposed, expedited workflow. A trained neural network predicts full-resolution surface morphology and current maps from undersampled inputs.
• Figure 3: A high-level diagram of our neural-network architecture. Our model, Sparse-C-AFM, can upsample surface morphology and current maps at multiple resolutions and levels of sparsity.
• Figure 4: Qualitative results for $\times$4 upsampling of an unseen MoS$_{2}$ sample. The top and bottom rows represent surface morphology and current channels from the C-AFM mapping, respectively. All results are visualized in RGB color-space using the matplotlib Python package set to the viridis color mapping scheme.
• Figure 5: Are SparseCAFM-upsampled current maps true to life? We compare the electrical characteristics (e.g., mean surface current, coverage percentage, etc) of sparsely sampled MoS$_2$ to SparseCAFM predictions at varying levels of sparsity ($\times 4$, $\times 16$, $\times 64$). The scorecard figure above reports the relative, mean absolute errors of SparseCAFM predictions and sparse current maps (baseline) using ground-truth data collected using classic-flow C-AFM scanning. For example, area extended shape values of SparseCAFM-upsampled current maps are over 60% more accurate than our baseline using a total data reduction factor of $\times 64$.