Building and deploying community nanotechnology software tools on nanoHUB.org - Atomistic simulations of multimillion-atom quantum dot nanostructures

The Network for Computational Nanotechnology (NCN) is a multi-university, NSF-funded initiative with a mission to lead in research, education, and outreach to students and professionals while at the same time deploying a unique web-based infrastructure to serve the nation's National Nanotechnology Initiative. The primary NCN outreach vehicle is our web site http://nanoHUB.org, which currently provides interactive on-line simulation and educational resources such as tutorials, seminars, and on-line courses. The educational and outreach resources were used in 2004 by over 3,200 users. Over 1,000 users performed over 60,000 on-line simulations. The raw web-page hit exceeded 3.7 million. Around 30 research codes are available ranging from toy-models to sophisticated simulation engines. The NCN provides the resource for models, simulation and computation without any software installation for users via web-delivery. All the NCN services are free of charge and reach a broad audience. One facet of this infrastructure involves the development of new "community codes" that provide the nanoscience research community with new capabilities and that lay a foundation for a new generation of CAD tools. Each research team has a research mission to move its field ahead and an equally important mission to contribute resources to the NCN's infrastructure. One such community tool is NEMO 3D which has been publicly released and recently expanded in its capabilities through detailed numerical performance analysis. The growth of self-assembled semiconductor quantum dots (QDs) is driven by strain, induced by the mismatch of the lattice constants of the QD material (in this work, InAs) and that of the barrier material (GaAs). The resulting long-range strain field strongly modifies the energy diagram of the system, and has to be accounted for in realistic simulations of QD electronic properties. The nanoelectronic modeling tool NEMO-3D is designed to provide quantitative estimates of QD-bound electron and hole states by treating the system on the atomistic level. It captures the strain by adjusting the positions of constituent atoms so as to minimize the total elastic energy computed in the VFF Keating model. The displacements of atoms from their bulk positions are incorporated into the 20-band nearest-neighbor sp3d5s* tight-binding Hamiltonian. A single dome-shaped InAs QD, with height of 1.8 nm and diameter of 19.2 nm, positioned on a 0.6-nm-thick wetting layer (WL), embedded in a GaAs barrier material is considered. Since strain is a long-range phenomenon, it has to be computed in a domain much larger than the quantum dot itself, while the domain for the electronic calculation can be smaller, since the electronic wave function is confined. Scaling to tens of millions of atoms: NEMO-3D uses the conjugate-gradient algorithm to find equilibrium positions of atoms, and the Lanczos algorithm to diagonalize the sparse Hamiltonian matrix. In order to extend the treatment to multi-million atom systems, both algorithms are parallelized using MPI. For example, on 16 64-bit Itanium2 CPUs with memory of 11.5 GB/CPU (NCSA Teragrid) it is possible to compute strain in 200-million-atom, and electronic structure of 35- million-atom devices. NEMO-3D exhibits a linear scaling, both of computational time and memory, as a function of the system size. This work presents a systematic study of the strain calculated within a domain consisting of up to 64 million atoms, followed by an electronic calculation on a subdomain containing up to 21 million atoms. Unique and targeted eigenstates can be extracted from system matrices of order 4×10 ⁸.