Study on Transformer-CNN based FDTD Method
- Resource Type
- Conference
- Authors
- Du, Shuxian; Sun, Yaxiu; Sun, Ruiying; Shan, Jingxin; Wang, Xiaoyang
- Source
- 2023 IEEE 11th Asia-Pacific Conference on Antennas and Propagation (APCAP) Antennas and Propagation (APCAP), 2023 IEEE 11th Asia-Pacific Conference on. volume1:1-2 Nov, 2023
- Subject
- Components, Circuits, Devices and Systems
Fields, Waves and Electromagnetics
Graphics processing units
Transformers
Electromagnetic fields
Time-domain analysis
Finite difference methods
- Language
This paper proposes a novel structure of Transformer-CNN to accomplish the computation process of 2D FDTD. Two CNNs are used to compute E and H respectively, and a transformer is used to accomplish time-marching step of FDTD. The structure is suitable for parallel computation, thus can accelerate the computation process of FDTD. Experiment shows that Transformer-CNN based FDTD uses less time than traditional FDTD to compute electromagnetic field in a $1\mathrm{m} \times1 \mathrm{m}$ computation domain.