APA
Click to copy
Jakhar, A. K., Basu, D., Dimple, K., Guglani, S., Dasgupta, A., & Roy, S. (2024). A Fast Metalearning Algorithm for Neural Network Enabled Uncertainty Quantification of Graphene Based Interconnects with Passive Shielding. 2024 IEEE International Symposium on Electromagnetic Compatibility, Signal &Amp; Power Integrity (EMC+SIPI).
Chicago/Turabian
Click to copy
Jakhar, Asha Kumari, Dyuti Basu, K. Dimple, Surila Guglani, Avirup Dasgupta, and Sourajeet Roy. “A Fast Metalearning Algorithm for Neural Network Enabled Uncertainty Quantification of Graphene Based Interconnects with Passive Shielding.” 2024 IEEE International Symposium on Electromagnetic Compatibility, Signal & Power Integrity (EMC+SIPI) (2024).
MLA
Click to copy
Jakhar, Asha Kumari, et al. “A Fast Metalearning Algorithm for Neural Network Enabled Uncertainty Quantification of Graphene Based Interconnects with Passive Shielding.” 2024 IEEE International Symposium on Electromagnetic Compatibility, Signal &Amp; Power Integrity (EMC+SIPI), 2024.
BibTeX Click to copy
@article{asha2024a,
title = {A Fast Metalearning Algorithm for Neural Network Enabled Uncertainty Quantification of Graphene Based Interconnects with Passive Shielding},
year = {2024},
journal = {2024 IEEE International Symposium on Electromagnetic Compatibility, Signal & Power Integrity (EMC+SIPI)},
author = {Jakhar, Asha Kumari and Basu, Dyuti and Dimple, K. and Guglani, Surila and Dasgupta, Avirup and Roy, Sourajeet}
}
Inserting passive shield lines in between the active and victim conductors has become a standard approach for mitigating the crosstalk effects in multi-walled carbon nanotube (MWCNT) and multilayer graphene nanoribbon (MLGNR) interconnect networks. However, the insertion of the shield lines augments the number of conductors of the network, thereby making the training of artificial neural network (ANN) surrogate models of such networks from SPICE simulations a time-consuming process. In this paper, this problem is addressed using a novel multistage metalearning algorithm. In the proposed algorithm, an additional ANN is trained using data extracted from SPICE simulations of the network where the active and victim lines are represented using the rigorous multiconductor circuit (MCC) model and the shield lines are represented using the compact but approximate equivalent single conductor (ESC) model. Finally, the metadata extracted from this additional ANN is leveraged to achieve higher training efficiency not possible using standard metalearning algorithms. Validation examples of MWCNT and MLGNR interconnect networks at 7-nm technology node are presented in this paper.