PERFORMANCE MODELING AND ANALYSIS OF GRAPHENE-BASED ON-CHIP VLSI INTERCONNECTS

Abstract

As per ITRS reports, the conventional on-chip interconnects will reach their performance limits at 8 nm technology node. It is due to the higher surface roughness scattering, grain boundary scattering, and electromigration effects. Further, scaling leads to the reduction in pitch and interconnects cross-sectional dimensions that result in increased capacitive effects and resistive effects. The higher operating speed requirements in state-of-the-art interconnects result in increased inductive effects. These increased effects of resistance, inductance, and capacitance in the conventional nanoregime interconnects severely impact their propagation delay and signal integrity performance parameters. The researchers are thus constantly looking for alternative materials that can serve as potential candidates to replace the conventional on-chip interconnects. The graphene-based interconnects such as graphene-nanoribbons (GNRs), carbon nanotubes (CNTs), and copper-carbon nanotubes (Cu-CNTs) have emerged as a probable solution to address the bottleneck faced by conventional Cu interconnects. This thesis presents the closed-form exponential matrix - rational approximation (EM-RA) model to capture the impact of edge-roughness variations on propagation delay and peak crosstalk in coupled-MLGNR interconnects. The proposed model is validated with the industry-standard SPICE simulator. It is observed that the obtained results lie within 1% error when compared with HSPICE simulations ensuring the excellent accuracy of the proposed model. The proposed closed-form model provides a speed-up of 75 and 51 for coupled-two and coupled-three MLGNR interconnects networks, respectively when the computational CPU time complexity is compared with HSPICE. The results for the crosstalk delay report lesser variations in wider MLGNRs (W > 45nm) compared to the narrow MLGNRs (W ≤ 45 nm). An eye height of 864.53 mV and an eye width of 220.17 ps is observed during the eye-diagram analysis of fully diffusive MLGNR lines.