Study the Temperature-Dependent Modeling for Performance Optimization ff Multilayer Graphene Nanoribbon (MLGNR) Based VLSI Interconnects

Abstract

The accurate performance of on-chip interconnects plays an important role to optimize the performance of integrated circuits (ICs) in deep sub-micron (DSM) technology nodes. Due to technology scaling, there has been a significant reduction in dimension size. Hence, problems of the mean free path for electrons, surface scattering from the boundaries of ultra-narrow conductors as well as grain boundary scattering inhibit electronic conduction in the copper wires to an unacceptable level. Due to technology scaling, thermal issues have also become a major challenging factor in the possible usage of on chip interconnect material for designing high performance integrated circuits. Consequently, alternative solutions such as graphene nano-ribbon (GNR) interconnects have been proposed in order to avoid the problems associated with global on-chip wires altogether. This thesis work includes the thermally aware circuit modeling and performance analysis of multilayer graphene nano-ribbon (MLGNR) based VLSI interconnects. The temperature-dependent performance in terms of propagation delay, power dissipation and crosstalk-induced voltage noise waveform at the far end of victim line, of MLGNR interconnects, have been analyzed at 22-nm technology node. SPICE simulations using PTM level 54 model were carried out to validate the findings. The results obtained through simulation are compared with conventionally used copper interconnects and it is observed that MLGNR outperforms its counterpart at different lengths of interconnects ranging from 200μm to 1000μm over a temperature range of 300K to 500K. A comparative performance analysis between MLGNR interconnects with resistance estimated using thermally aware model and temperature independent model (conventional) is investigated. Average relative improvements of 37.24% and 26.34% in propagation delay and power dissipation respectively are achieved using a thermally aware model in comparison with a temperature independent model of MLGNR resistance, with length variations from 200μm to 1000μm. Further, an average relative improvement in the time duration reduction of victim output, for the same range of interconnect lengths, is achieved about 35% by using a thermally aware model instead iv of a temperature independent model of MLGNR resistance. Obtained results reflect that the thermally aware modeling of MLGNR is important for its performance optimization in DSM technology nodes. After the temperature-dependent comparative performance analysis, MLGNR comes out as promising alternative to copper for the use as future VLSI interconnects, due to its less sensitivity with temperature dependent scattering.