Characterization of Phonon Dynamics and Thermal Environments in FinFET Architectures

Physical device scaling for traditional chip architectures is expected to end by 2021. As a result, alternatives are being aggressively pursued that can meet the increasing computing demands imposed by data-centric computing and sensor networks. Three-dimensional IC architectures are particularly promising due to high levels of integration and enhanced performance. However, increased integration leads to thermal bottlenecks that greatly reduce device performance. To understand the bottlenecks, we use atomistic modeling techniques to characterize the nanoscale phonon dynamics and map the thermal landscape of modern fin field-effect transistors (FinFET). We identify phonon modes of FinFET subcomponents that contribute strongly to thermal transport using a lattice dynamics (LD) framework based on a quasi-harmonic Green-Kubo approximation. In parallel, we generate subcomponent temperature profiles using molecular dynamics (MD) simulations. We establish a relationship between the MD temperature profiles and the corresponding populations of the LD-identified phonon modes using a machine learning algorithm. This relationship is used to predict the full phonon dynamics of the entire FinFET device.