Device and method for electromagnetic field simulation

Patent metadata

Field	Value
Publication number	US-11210561-B2
Application number	US-201916718720-A
Country	US
Kind code	B2
Filing date	Dec 18, 2019
Priority date	Dec 18, 2018
Publication date	Dec 28, 2021
Grant date	Dec 28, 2021

Abstract

A device and a method for electromagnetic field simulation are provided. The image processing device is to obtain, with a first resolution, a first electromagnetic field simulation result of a simulation region, and to derive a second electromagnetic field simulation result with a second resolution, according to the first electromagnetic field simulation result, using a model trained based on a deep learning method. The second resolution is higher than the first resolution.

First claim

The invention claimed is:

1. A device, comprising:
a processor to couple to a memory to,
obtain, with a first resolution, a first electromagnetic field simulation result of a simulation region; and
derive a second electromagnetic field simulation result with a second resolution higher than the first resolution, according to the first electromagnetic field simulation result, using a model trained based on a deep learning method,
wherein the first electromagnetic field simulation result and the second electromagnetic field simulation result comprise a four-dimensional data set, and
to derive the second electromagnetic filed simulation result, the processor is to convert the first electromagnetic field simulation result into the second electromagnetic filed simulation result by,
decomposing the four-dimensional data set into a plurality of two-dimensional data sets;
for each two-dimensional data set of the plurality of two-dimensional data sets, improving a resolution of the two-dimensional data set using the model; and
combining a plurality of the improved two-dimensional data sets, whose resolutions have been improved, into an improved four-dimensional data set, as the second electromagnetic filed simulation result with the second resolution.

2. The device according to claim 1 , wherein the processor is to obtain the first electromagnetic field simulation result using a finite difference time domain (FDTD) method.

3. The device according to claim 2 , wherein the processor is to calculate the first electromagnetic field simulation result based at least on an electromagnetic wave propagation frequency range in the simulation region, geometric parameters and material parameters of the simulation region, and a mesh size.

4. The device according to claim 1 , wherein the deep learning method includes a convolutional neural network (CNN) method.

5. The device according to claim 1 , wherein the deep learning method comprises a super-resolution generative adversarial network (SRGAN).

6. The device according to claim 1 , wherein the processor is to obtain the model by:
initializing an electromagnetic field simulation by using the first electromagnetic field simulation result and the second electromagnetic field simulation result as training data, respectively; and
deriving the model based on the training data using the deep learning method.

7. The device according to claim 6 , wherein to obtain the training data, the processor is to obtain the electromagnetic field simulation result of the simulation region using a finite difference time domain (FDTD) method, based on a first mesh size corresponding to the first resolution and a second mesh size corresponding to the second resolution, respectively.

8. The device according to claim 6 , wherein the deep learning method comprises a super-resolution generative adversarial network (SRGAN), and to obtain the model the processor is to obtain a generator network and a discriminator network of the SRGAN using the training data, and use the generator network as the model.

9. The device according to claim 1 , wherein the model comprises a plurality of models, which are used for mapping between a plurality of first resolutions and/or a plurality of second resolutions, respectively.

10. A method for electromagnetic field simulation, comprising:
by a processor that couples to a memory,
obtaining, with a first resolution, a first electromagnetic field simulation result of a simulation region; and
deriving a second electromagnetic field simulation result with a second resolution that is higher than the first resolution, according to the first electromagnetic field simulation result, using a model trained based on a deep learning method,
wherein the first electromagnetic field simulation result and the second electromagnetic field simulation result comprise a four-dimensional data set, and
the deriving the second electromagnetic filed simulation includes converting the first electromagnetic field simulation result into the second electromagnetic filed simulation result by,
decomposing the four-dimensional data set into a plurality of two-dimensional data sets;
for each two-dimensional data set of the plurality of two-dimensional data sets, improving a resolution of the two-dimensional data set using the model; and
combining a plurality of the improved two-dimensional data sets, whose resolutions have been improved, into an improved four-dimensional data set, as the second electromagnetic filed simulation result with the second resolution.

11. The method according to claim 10 , wherein the first electromagnetic field simulation result is obtained using a finite difference time domain (FDTD) method.

12. The method according to claim 11 , wherein the first electromagnetic field simulation result is calculated based at least on an electromagnetic wave propagation frequency range in the simulation region, geometric parameters and material parameters of the simulation region, and a mesh size.

13. The method according to claim 10 , wherein the deep learning method comprises a super-resolution generative adversarial network (SRGAN).

14. The method according to claim 10 , wherein the model is obtained by:
initializing the electromagnetic field simulation by using the first electromagnetic field simulation result and the second electromagnetic field simulation result as training data, respectively; and
deriving the model based on the training data using the deep learning method.

15. The method according to claim 14 , wherein to obtain the training data, the electromagnetic field simulation result of the simulation region is obtained using a finite difference time domain (FDTD) method, based on a first mesh size corresponding to the first resolution and a second mesh size corresponding to the second resolution, respectively.

16. The method according to claim 14 , wherein the deep learning method comprises a super-resolution generative adversarial network (SRGAN), and a generator network and a discriminator network of the SRGAN are obtained using the training data, and the generator network is used as the model.

Assignees

• Fujitsu Ltd

Inventors

• Zhou Rong
• Wang Liuan
• Sun Jun

Classifications

• G06F30/23 Primary

  using finite element methods [FEM] or finite difference methods [FDM] · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06T3/4046

  using neural networks · CPC title

• G06F30/27

  using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model · CPC title

• G06K9/6256 Primary

  Physics · mapped topic