Here is the abstract you requested from the Thermal_2017 technical program page. This is the original abstract submitted by the author. Any changes to the technical content of the final manuscript published by IMAPS or the presentation that is given during the event is done by the author, not IMAPS.
|First-principles 3D thermal simulation of microscale devices|
|Keywords: 3D simulation, multilayer devices, first principles|
|Contemporary electronics pose significant thermal challenges that go well beyond geometric (Moore-driven) increases in dissipated power density. Heat removal from small sources or across thin films, both common occurrences in semiconductor technologies, is less efficient than classically predicted due to inherent microscopic dynamics. These effects are captured in detail by the Boltzmann transport equation (BTE), but most solutions are limited to 1D transport within a single material and thus unable to describe multidimensional heat spreading crucial to accurate system-level modeling. Here, we present direct 3D microscopic thermal simulations of realistic device geometries using deviational Monte Carlo techniques. These allow fully parallelised numerical solution of the BTE from stochastic trajectories of mutually independent simulation particles. Each material is treated from first principles, i.e. the fundamental inputs that govern the random energy motion are derived directly from the atomistic crystal properties. We illustrate the capabilities through finFET and LED solvers developed for our open-source simulation software almaBTE. FinFET technology realises transistors atop 10-20nm narrow 'fins' that protrude 50-100nm from the Si substrate. Accurate thermal modeling becomes instrumental to the design process because junction temperatures in an active 'heater' device can only be inferred indirectly from readings on an adjacent 'sensor' device and the theoretical heater-sensor thermal cross-talk. Conventional diffusive theory underpredicts the measured sensor temperatures by up to sixfold while our parameter-free simulations match within 20%. Our LED solver targets commercially used 5-layer cylindrical structures. The dissipation profile can either be prescribed manually or taken directly from simulated electron-hole recombination rates. Simulations of a GaAs/AlGaAs device with 4 micron thickness and radius revealed junction-to-case resistances to exceed conventional predictions by 60%. Both case studies highlight the recent advances in, and growing need for, multidimensional 'beyond Fourier' thermal simulation of microscale devices.|
|Bjorn Vermeersch, Research Engineer