Application of Artificial Neural Network Surrogate Models for Efficient Signal Integrity Analysis in Emerging Graphene-Based Interconnects

Abstract — With advancing technology nodes, conventional copper on-chip interconnects are becoming more susceptible to different scattering mechanisms such as sidewall scattering, surface roughness scattering and grain boundary scattering. These scattering mechanisms increase the per-unit-length resistance of the interconnects, thereby leading to increased signal attenuation, latency, and power losses. In order to address these limitations of conventional copper interconnects, more advanced interconnect technologies such as carbon nanotubes and hybrid copper-graphene interconnects are currently being investigated. These newer technologies exploit the enhanced electrical and material properties of novel 2D materials such as graphene to improve the overall conductive properties of the interconnects. However, in order to model the electrical properties of graphene-based interconnects, highly complicated equivalent circuit models are required, the solution of which are extremely time consuming. One approach to mitigate the high computational costs of modeling such novel interconnects is by using artificial neural network models. In this article, we review the current state-of-the-art in artificial neural networks to efficiently model the transient responses of the aforementioned emerging graphene-based interconnects.