Power-delay-product (PDP) optimal repeater size and number design for horizontal and vertical multi-layer graphene nanoribbon (MLGNR) interconnects is implemented. Horizontal MLGNR (HMLGNR) interconnects, both top-contacted (TC) and side-contacted (SC), and vertical MLGNR (VMLGNR) interconnects are considered. The four most common optimization algorithms including ant colony optimization for continuous domains (ACOR), particle swarm optimization (PSO), artificial bee colony (ABC), and gray wolf optimization (GWO) are compared in terms of speed and accuracy to choose the best suited algorithm for optimizing repeater size and number in MLGNR interconnects. Edge backscattering probability, surface roughness amplitude, and doping concentration, due to their significant effect on the performance of MLGNR interconnects are considered as input parameters to optimization algorithms. The ACOR optimization results are then utilized to train a back-propagation neural network that performs the optimal repeater design very much faster than optimization algorithms. It is shown that VMLGNR or HMLGNR interconnects may need less repeater number depending on the edge backscattering probability, surface roughness amplitude, and doping concentration. HMLGNR interconnects with top contacts always need less repeater size than other interconnect types irrespective of edge backscattering probability, surface roughness amplitude, and doping concentration.
