The Search for Alternative Computational Paradigms

Nanometer processes are characterized by extremes of process variations, noise, soft errors, and other nonidealities, which threaten to nullify the intrinsic benefits of scaling. The resulting robustness and energy efficiency problem cannot be addressed in a cost-effective manner solely through advances in manufacturing. Alternative models of computation are needed that thrive in the presence of statistical variations in the underlying device and circuit fabric. This article explores communications-inspired models of computation supported by innovative robust circuit and logic fabric design approaches. These models share the common feature of leveraging dense networks with information exchange and coupling among nodes to enhance robustness without compromising energy efficiency. Promising post-silicon devices such as carbon nanotubes (CNTs) offer an attractive platform on which to build such computational systems. This article identifies opportunities and challenges in designing robust and low-power SoCs in emerging nanoscale process technologies, employing radically new modes of computation.