NERSC Initiative for Scientific Exploration (NISE) 2011 Awards
Simulation of Inelastic Decoherence in Electronic Transport Through Nanoscale Structures
Sayeef Salahuddin, University of California - Berkeley
Associated NERSC Project: Massively parallel quantum transport simulation of nano scale electronic devices for ultra low power computing (m946)
|NISE Award:||750,000 Hours|
|Award Date:||March 2011|
Inelastic scattering, where an electron can gain or loose energy, significantly affects the way electrons move through a nanostructure. Many experiments show that inelastic scattering presents a serious bottleneck in the efficiency for solar cells and thermoelectric devices. However, modeling inelastic scattering properly in a nanostructure while also capturing the quantum effects has been intractable due to significant numerical burden. In this research, we shall attempt to use physical symmetry of the problem and numerical algorithms to massively parallelize the simulation methodology so that simulation of decoherence in real structures can eventually be performed.
Inelastic scattering significantly affects the way electrons move through a nanostructure. Many experiments show that inelastic scattering presents a serious bottleneck in the efficiency for solar cells and thermoelectric devices. However, modeling inelastic scattering properly in a nanostructure presents significant challenge. This is because, within a full quantum transport description, which is necessary to account for quantum effects in a nanostructure, an inelastic scattering couples all the energy levels in the system. In addition, the broadening of states that ensue from such decoherence needs to be solved self-consistently with transport. This is why simulation of atomistic inelastic decoherence in electronic transport for realistic devices is often considered to be impossible.
We propose to massively parallelize the decoherence calculation within this proposal so that realistic structures can eventually be simulated. The key idea stems from the fact that, only a few phonon/photon frequencies are coupled through inelastic decoherence. Therefore, we shall try to use the OpenMP to take advantage of the multi-core nodes where each node will take care of the neighboring energy states in the calculation. A second way the parallelization process will work is to take advantage of the need for self-consistency. All the relevant energies will be parallelized to individual nodes where each node will parallelize the coupling between neighboring frequencies. Additionally, each iteration will use the previous iteration as an input.
In summary, this effort, if successful, will enable simulation of electronic transport in presence of inelastic decoherence in realistic structures beyond the state of the art. Since such decoherence mechanism is critical to the physics of all sorts of solar energy conversion and thermoelectric devices, the practical implication could be significant. From an implementation point of view, this work will (i) explore new physics and numerical algorithms (ii) investigate if OpenMP could enhance the performance of such simulations at the node level and (iii) also enhance the performance of an existing code developed with the support of a previous NISE award. Notably, this code currently has a scaling over 8192 cores. We have already used this code to perform state of the art simulations that generated multiple publications in 2010 including an issue cover story from Applied Physics Letters. If the OpenMP implementation proposed in this request is successful, we expect the scaling to improve at least by one order of magnitude.