Noise-Aware Uncertainty Quantification of MLGNR Interconnects using Fast Trained Artificial Neural Networks

This paper presents a prior knowledge accelerated transfer learning technique to efficiently train artificial neural networks (ANNs) to address uncertainty in on-chip multilayer graphene nanoribbon (MLGNR) interconnects. The salient feature of the proposed training technique is that it accounts for the variation in the amplitude, frequency, and phase angle of the power supply noise and ground bounce signals. Consequently, the resultant ANN model is noise-aware and can be used to anticipate the eye diagram attributes of the MLGNR interconnects for any arbitrary noise signal without the need for any retraining.