SparseC-AFM: fast 2D-material acquisition & analysis with super resolution models

---

This is the official Pytorch implementation of our paper: SparseC-AFM: a deep learning method for fast and accurate characterization of MoS₂ with C-AFM. We present a novel method for rapid acquisition and analysis of C-AFM scans using a super-resolution model based on the work of SwinIR. In this repository, you can find the datasets and model weights used in our paper, as well as scripts to train and deploy our model on your own datasets.

Below we include our enviornments, data, and model weights.

Getting Started

We use anaconda for all Python enviornment management. Clone our enviornment using the command below.

conda env create -f environment.yml

Once installed, activate the enviornment.

conda activate sparse-cafm

Datasets

Path	Material	Height Maps	Current Maps	Substrate	Mode	# Samples	# Data Points	Resolutions
data/raw-data/3-12-25	BTO	✅	❌	---	Tapping (AFM Only)	4	16	{64, 128, 256, 512}
data/raw-data/2-6-25	MoS2	✅	✅	SiO2-Si	Contact	1	4	{64, 128, 256, 512}
data/raw-data/1-23-25	MoS2	✅	✅	SiO2-Si	Contact	1	5	{512}
data/raw-data/11-19-24	MoS2	✅	✅	SiO2-Si, Sapphire	Contact	2	10	{512}

Model Weights

Path	Upscaling Factor	Material	Height Maps	Current Maps	Substrates
data/weights/...2x.pth	$$\times2$$	MoS2	✅	✅	SiO2-Si, Sapphire
data/weights/...4x.pth	$$\times4$$	MoS2	✅	✅	SiO2-Si, Sapphire
data/weights/...8x.pth	$$\times8$$	MoS2	✅	✅	SiO2-Si, Sapphire

Citation

@article{harris2025sparsec-afm,
  title={SparseC-AFM: a deep learning method for fast and accurate characterization of MoS2 with C-AFM},
  author={Harris, Hossain, Qui, Zhang, Ma, Chen, Gu, Tongay, Celano},
  conference={SPIE},
  year={2025}
}

License

We release our work under the Apache License 2.0 ❤️.