CMAS Lab - Combining Prior Knowledge With Transfer Learning (PKID-TL)

Journal article

Surila Guglani, Asha Kumari Jakhar, A. Dasgupta, Sourajeet Roy
IEEE Transactions on Components, Packaging, and Manufacturing Technology, 2025

Semantic Scholar DOI

Cite

APA Click to copy
Guglani, S., Jakhar, A. K., Dasgupta, A., & Roy, S. (2025). Combining Prior Knowledge With Transfer Learning (PKID-TL) for Fast Neural Network Enabled Uncertainty Quantification of Graphene On-Chip Interconnects. IEEE Transactions on Components, Packaging, and Manufacturing Technology.

Chicago/Turabian Click to copy
Guglani, Surila, Asha Kumari Jakhar, A. Dasgupta, and Sourajeet Roy. “Combining Prior Knowledge With Transfer Learning (PKID-TL) for Fast Neural Network Enabled Uncertainty Quantification of Graphene On-Chip Interconnects.” IEEE Transactions on Components, Packaging, and Manufacturing Technology (2025).

MLA Click to copy
Guglani, Surila, et al. “Combining Prior Knowledge With Transfer Learning (PKID-TL) for Fast Neural Network Enabled Uncertainty Quantification of Graphene On-Chip Interconnects.” IEEE Transactions on Components, Packaging, and Manufacturing Technology, 2025.

BibTeX Click to copy

@article{surila2025a,
  title = {Combining Prior Knowledge With Transfer Learning (PKID-TL) for Fast Neural Network Enabled Uncertainty Quantification of Graphene On-Chip Interconnects},
  year = {2025},
  journal = {IEEE Transactions on Components, Packaging, and Manufacturing Technology},
  author = {Guglani, Surila and Jakhar, Asha Kumari and Dasgupta, A. and Roy, Sourajeet}
}

Abstract

In this article, an artificial neural network (ANN)-enabled uncertainty quantification (UQ) technique is developed for graphene on-chip interconnects. In the proposed technique, primary ANNs are trained to emulate the signal integrity (SI) characteristics of multiwalled carbon nanotube (MWCNT) and multilayer graphene nanoribbon (MLGNR) interconnects. The training of the primary ANNs is accelerated by using information elicited from other ANNs, known as secondary ANNs, that have been pretrained to perform related tasks. In this article, the elicited information takes two forms: 1) the optimized values of the weights and bias terms and 2) the output features of the secondary ANNs. Algorithms to intelligently infuse these two distinct types of information using a multistep training process have been proposed to ensure the best possible speedup. Validation examples spanning both MWCNT and MLGNR interconnect networks and different technology nodes have been presented.