Atoms to Devices: Predicting Thermal Properties of Materials in Technological Applications

Recent advances of atomistic modeling techniques have shown remarkable accuracy while predicting heat transport mechanisms in isolated nanoscale systems, such as thin films with confining surfaces, single solid-solid interfaces between materials with different degrees of crystallinity or disorder. Heat is mostly carried by phonons in semiconductors and insulators. When the system size becomes comparable to the characteristic lengths of phonons, heat conduction is affected by phonon confinement and enhancement of scattering due to boundaries, interfaces, and various defects. Phonon confinement results in localization of modes in spatial regions. The effect of scattering mechanisms on phonon transport properties could also be highly varied. The thermal transport properties of nanoscale systems that include multiple confining surfaces, interfaces, and materials with different degrees of crystallinity or disorder, is far from understood. The computational costs of atomistic approaches prohibit us to explore the complete structural parameter space of large systems, that influence phonon transport. The state-of-the-art (SoA) deeply scaled sub-10 nanometers transistors are heterogeneous structures that include nm-scale crystalline semiconductors, disordered dielectric materials and metals, within confining interfaces. The confined geometry results in a net increase of the power density within the structures. The associated self-heating accelerates the defect generation mechanisms leading to decline of performance and reliability issues. The SoA thermal models of microelectronic circuits utilize continuum level models to predict the temperature distribution in the structures, even after 40+ years of being declared outdated. The models also incorporate input from non-continuum models that are better suited to describe the physics at smaller length scales. However, the transistor-level thermal properties are often approximated from the bulk thermal properties or computed using effective models. The effective models include several parameters that need to be calibrated using measurement data. The calibrated parameters may not even accurately represent behavior in dynamic operating conditions. The heat transfer in nm-scale systems can be drastically different from their bulk counterparts. The effective models does not include these details and as a result, cannot explain thermal bottlenecks or runaways or reveal any mitigation strategies. A predictive thermal model must consider the effects of confinement to accurately predict temperatures and identify thermal bottlenecks. In the past decades, algorithmic improvements, and dramatic increase of computing power, have enabled rigorous techniques such as molecular dynamics (MD) simulations to accurately model heat transport in finite systems, with disordered components and interfaces. MD simulations implicitly include the full dynamics of anharmonic interactions and do not make any assumptions about scattering mechanisms. In this talk, I will discuss how we use atomistic modeling approaches combined with machine learning methods to predict thermal properties of sub-10-nm confined geometry field effect transistors and other complex devices.