SemiConLens: Visual Analytics for 2D Semiconductor Discovery
New tool combines ML and human expertise to find better semiconductors...
SemiConLens, developed by Kavinda Athapaththu and collaborators, addresses critical bottlenecks in 2D semiconductor discovery. Traditional approaches like Density Functional Theory (DFT) or standard machine learning struggle with small datasets and reliability issues. SemiConLens introduces a Correlation Aware Multivariate Imputation (CAMI) method combined with autoencoder models to better learn from limited data while revealing prediction uncertainty. This allows researchers to handle sparse datasets more effectively.
On top of this, the tool provides an interactive visualization module with three linked views. A novel circular glyph design displays multiple key attributes and uncertainty levels per candidate, while a cluster-aware layout optimization helps users group and compare similar materials. Quantitative evaluations, expert interviews, and use cases confirm that SemiConLens enables material scientists to efficiently filter, discover, and validate promising 2D semiconductors, making the discovery process more reliable and trustable.
- Introduces CAMI method and autoencoders to handle small and sparse datasets in semiconductor discovery.
- Visualization uses circular glyphs and cluster-aware layout to compare candidates and uncertainty.
- Validated through expert interviews, quantitative evaluations, and use cases in materials science.
Why It Matters
Speeds up reliable discovery of 2D semiconductors, crucial for next-gen electronics beyond silicon.