Predictive atomistic simulations of electronic properties of realistic nanoscale devices: A multiscale modeling approach.

摘要

Scaling of CMOS towards its ultimate limits, where quantum effects and atomistic variability due to fabrication, along with recent emphasis on heterogeneous integration of non-digital devices for increasing the functional diversification presents us with fundamentally new challenges. A comprehensive understanding of design and operation of these nanoscale transistors, and other electronic devices like RF-MEMS, requires an insight into their electronic and mechanical properties that are strongly influenced by underlying atomic structure. Hence, continuum descriptions of materials and use of empirical models at these scales become questionable. This increase in complexity of electronic devices necessitates an understanding at a more fundamental level to accurately predict the performance and reliability of these devices. The objective of this thesis is to outline the application of multiscale predictive modeling methods, ranging from atoms to devices, for addressing these challenges. This capability is demonstrated using two examples: characterization of (i) dielectric charging in RF-MEMS, and (ii) transport properties of Ge-nanofins. For characterizing the dielectric charging phenomenon, a continuum dielectric charging model, augmented by first principles informed trap distributions, is used to predict current transient measurements across a broad range of voltages and temperatures. These simulations demonstrate using ab initio informed model not only reduces the empiricism (number of adjustable parameters) in the model but also leads to a more accurate model over a broad range of operating conditions, and enable the precise determination of additional material parameters. These atomistic calculations also provide detailed information about the nature of charge traps and their trapping mechanisms that are not accessible experimentally; such information could prove invaluable in defect engineering. The second problem addresses the effect of the in-homogeneous strain profiles, inherent in strain-engineered Ge nanofins, on their transport properties. Fully atomistic simulations, involving a combination molecular dynamics simulations with first-principles based force-fields and semi-empirical tight binding calculations, coupled with linearized Boltzmann model are used to calculate the hole transport properties of realistic Ge nanofins (heights 5-15nm and widths 5nm-40nm). Our simulations predict the technological limit of phonon limited hole mobility improvement in Ge channel PMOS devices (H<15nm) and present geometric guidelines for patterning nanofins to engineer high performance uniaxial devices conducive to the existing top-down fabrication approaches. From these calculations, we demonstrate that realistic modeling of the devices requires a reduction in the empiricism of fitting parameters and incorporation of new multi-scale, multi-resolution approach spanning across various spatial and temporal scales. Such physics based predictive multiscale models facilitate an integrated approach for rapid development and pave the way for designing new advanced materials and devices.