Related Experiment Video
Updated: Sep 17, 2025
Advanced Experimental Methods for Low-temperature Magnetotransport Measurement of Novel Materials
Published on: January 21, 2016
Accelerating Low-k Dielectric Material Discovery: From Graph Machine Learning to Synthesis.
Zhao-Chen Xi1, Xin Wang1, Chang-Hao Wang1
1Multifunctional Materials and Structures, Key Laboratory of the Ministry of Education & International Center for Dielectric Research, School of Electronic Science and Engineering, Xi'an Jiaotong University, Xi'an 710049, P.R. China.
A new graph neural network (Res-GCN) model accelerates the discovery of low-permittivity dielectric materials. This machine learning approach significantly improves accuracy and reduces experimental time for finding novel materials for electronics.
Area of Science:
- Materials Science
- Computational Materials Science
- Machine Learning
Background:
- Discovering dielectric materials with specific permittivity is challenging, often relying on inefficient trial-and-error methods.
- Low-permittivity materials are crucial for semiconductor interlayers and communication substrates to reduce parasitic capacitance and transmission latency.
Purpose of the Study:
- To develop an efficient and scalable machine learning framework for predicting dielectric permittivity.
- To accelerate the discovery and synthesis of novel low-permittivity dielectric materials.
Main Methods:
- A graph neural network (Res-GCN) model was developed to predict permittivity directly from atomic structure.
- A material searching pipeline based on pattern recognition was integrated with the Res-GCN model.
- High-throughput screening of approximately 6000 material entries was performed.
Main Results:
- The Res-GCN model demonstrated significant accuracy improvements over classical and other deep learning models, with a root-mean-squared error of ~1.788.
- High-throughput screening identified promising low-permittivity candidates within minutes.
- Two novel dielectric ceramics with desired properties were successfully synthesized through only eight targeted experiments.
Conclusions:
- Machine learning, specifically graph neural networks, offers an efficient and scalable approach to accelerate dielectric material discovery.
- The developed framework significantly shortens the research cycle by enhancing the synergy between computational prediction and experimental validation.
