Towards extreme-scale coupled electrothermal simulations of realistic nano-devices
The goal of this project is to accelerate the simulation of realistic nano-devices with coupled electrothermal transport using advanced numerical methods on massively parallel architectures and a hierarchical modeling approach. The increase in heat dissipation and power consumption is currently reaching a critical level in integrated circuits. If this trend does not stop in the near future, it will no longer be possible to sufficiently cool down electronic devices, thus severely degrading their performance and lifetime. This problem is becoming even more important now that the size of the transistors, the active components of ICs, does not exceed a few nanometers and their gate contact lets more and more electrons leak between source and drain in stand by mode. Designing, fabricating, and characterizing novel energy-efficient nano-transistors is a long and expensive process that can be greatly supported by device simulation. This research activity can help reduce heat generation and manage power dissipation in nanostructures, provided that a technology computer aided design tool capable of treating both electrical and thermal transport at a quantum mechanical level and with an atomistic resolution exists. Drift-diffusion solvers, as widely used in industry and academia, usually offer energy-balance and electrothermal models that are computationally very efficient, but lack predictability at the nanometer scale where energy quantization, geometrical confinement, and intra- and band-to-band tunneling play a significant role. A quantum transport simulator can deliver accurate performance predictions at the nanoscale if electron-phonon and phonon-phonon scattering are taken into account so that energy gain/loss, anharmonic phonon decay, and self-heating effects are correctly described. The problems with such an approach are that it is computationally very intensive, it needs a calibration of its scattering mechanisms to compensate for the necessary approximations it contains, and it cannot be used to explore a large design space due to its limited computational efficiency. Here, we propose to address all these issues. The TORNAD project will engage in research along two directions to provide the device modeling community and the semiconductor industry with a beyond state-of-the-art quantum-transport simulation approach dedicated to next-generation nano-transistors such as 3-D Si, Ge, InGaAs nanowire and 2-D transistors: 1. massively parallel and advanced sparse numerical linear algebra algorithms will be developed and implemented to solve the non-equilibrium Green's function equations in the presence of dissipative scattering mechanisms. They will be benchmarked against existing techniques; 2. a hierarchical modeling approach going from density-functional theory down to quantum drift-diffusion calculations will be established. As a central point, the energy-balance and electrothermal model of a commercial drift-diffusion simulator, Sentaurus-Device , will be calibrated with results from an accurate, but computationally very intensive quantum transport solver.