Channel Inverse Design Using Tandem Neural Network
- Resource Type
- Conference
- Authors
- Ma, Hanzhi; Li, Er-Ping; Wang, Yuechen; Shi, Bobi; Schutt-Aine, Jose; Cangellaris, Andreas; Chen, Xu
- Source
- 2022 IEEE 26th Workshop on Signal and Power Integrity (SPI) Signal and Power Integrity (SPI), 2022 IEEE 26th Workshop on. :1-3 May, 2022
- Subject
- Components, Circuits, Devices and Systems
Engineered Materials, Dielectrics and Plasmas
Fields, Waves and Electromagnetics
Photonics and Electrooptics
Training
Performance evaluation
Conferences
Neurons
Artificial neural networks
Attenuation
Numerical models
high-speed link
inverse design
neural network
impedance and attenuation
- Language
A tandem neural network (NN) with R 2 score-based loss function is proposed in this paper for channel inverse design. Tandem NN consists of an inverse neural network from target performance to design parameters and a pre-trained forward neural network from design parameters to design targets. The training of the actual INN uses the fixed pre-trained forward model to evaluate the inverse design output. A channel inverse design example for target impedance and attenuation at multiple frequency points is applied in this paper to evaluate the performance of tandem NN. Numerical results show that tandem NN achieves a good design result compared with target performance and regular NN.