The award-winning results, presented at SC11, revealed the electron states of silicon nanowires, which have attracted attention as a core material for next-generation semiconductors. To verify the computational performance of the K computer, quantum-mechanical computations were performed on the electron states of a nanowire with approximately 100,000 atoms (20 nanometers in diameter and 6 nanometers long), close to the actual size of the materials, and achieved execution performance of 3.08 petaflops (representing execution efficiency of 43.6%). The results of the detailed calculations on the electron states of silicon nanowires, comprised of 10,000 to 40,000 atoms, clarified that electron transport characteristics will change depending on the cross-sectional shape of the nanowire.
The award marked the first time that a research group from Japan won the Gordon Bell Prize for Peak-Performance since 2004, when a team led by Professor Akira Kageyama, currently based at Kobe University, won the prize for performing a simulation of a geomagnetic dynamo using the first-generation Earth Simulator installed at the Japan Agency for Marine-Earth Science and Technology (JAMSTEC).
The application used in this current research is one of the core applications developed as part of the "Grand Challenges" programme - which, in the field of nanotechnology, is led by the Institute for Molecular Science at Japan's National Institute of Natural Sciences - of the Ministry of Education, Culture, Sports, Science and Technology (MEXT).
As a part of the High-Performance Computing Infrastructure (HPCI) initiative led by MEXT, both RIKEN and Fujitsu have been working together to develop the K computer and high-speed communications software (MPI library) for use on the K computer, with the aim of beginning shared use by November 2012.
Since the start of the "Development of a Next-Generation Supercomputer" project - now renamed "HPCI Initiative" - in 2006, RIKEN, the University of Tsukuba, and the University of Tokyo have been working to develop optimal calculation methods and programmes to verify the performance of the K computer using real applications. Researchers of computational science, whose aim is to elucidate natural phenomena using computer simulations, and researchers of computer science, whose aim is to drive technological advances in the application of supercomputers, have been working together to efficiently develop massively parallel code or RSDFT (Real-Space Density Functional Theory) based on density functional theory by discretizing it in real space, which is used to compute the energy states and electron states of matter. In the current research, the K computer was used to compute the electron states of silicon nanowires, ultrafine traces made of silicon atoms that have potential for use as a next-generation semiconductor material that generates rapid switching speeds and low current leakage.
In results computed using approximately two-thirds of the K computer system's capacity, execution performance of 3.08 petaflops was achieved. The research team succeeded in calculating the electron state of a silicon nanowire measuring 10 nm in diameter and 10 nm in length comprised of 39,696 atoms. These results were presented at the SC11 being held November 12-18 in Seattle, USA. The findings won the ACM Gordon Bell Prize for Peak-Performance in a genuine application programme.
The Gordon Bell Prize was established in 1987 by noted American computer scientist Gordon Bell to promote the development of parallel computing technology, and it is administered by the Association for Computing Machinery (ACM). Every year, the prize is awarded to the group that has applied parallel computing to practical scientific problems and produced outstanding results.
From among five finalist papers that had passed the first round of selections, the judges for this year's Gordon Bell Prize selected the paper "First-principles calculations of electron states of a silicon nanowire with 100,000 atoms on the K computer" for the Peak-Performance Prize. The Peak-Performance Prize is given for research that shows high real-world performance in a practical application, and underscores both the practical value of this research and the K computer's practical effectiveness.
The silicon nanowires that are the subject of this research, which serve as electron channels in next-generation field-effect transistors, comprise 10,000 to 100,000 silicon atoms and measure 1020 nm in diameter and roughly 10 nm in length. These results represent the world's first attempt at calculating the electron state, using quantum theory, for a 107,292-atom nanowire measuring 20 nm in diameter and 6 nm in length. Execution performance of 3.08 petaflops was achieved (representing execution efficiency of 43.6%), showing that high-precision simulations are possible at the nanoscale level that will be used in future devices. This work also examined the electron states of nanowires comprising 10,000 to 40,000 atoms, and measuring approximately 10 nm, but in different cross-sectional shapes (round, oblong, and dumbbell-shaped), and found that electron-transport characteristics were highly dependent on the nanoscale shape, showing that computational material sciences using the K computer can contribute greatly to nanoscale design decisions.
As the real applications running on the K computer received a very high evaluation, the research group was awarded the Gordon Bell Prize for Peak-Performance. This is significant in that it shows the international high performance computing (HPC) community has recognized the K computer's potential to advance science and technology. While continuing to gain a deeper understanding of silicon nanowires, researchers will also use these research experiences to raise the level of sophistication of other applications so that the K computer can bring results faster to a wider range of scientific fields.
"First-principles calculations of electron states of a silicon nanowire with 100,000 atoms on the K computer" is available at http://dl.acm.org/citation.cfm?id=2063386 . The authors are Yukihiro Hasegawa (RIKEN), Junichi Iwata (The University of Tokyo), Miwako Tsuji (University of Tsukuba), Daisuke Takahashi (University of Tsukuba), Atsushi Oshiyama (The University of Tokyo), Kazuo Minami (RIKEN), Taisuke Boku (University of Tsukuba), Fumiyoshi Shoji (RIKEN), Atsuya Uno (RIKEN), Motoyoshi Kurokawa (RIKEN), Hikaru Inoue (Fujitsu), Ikuo Miyoshi (Fujitsu), and Mitsuo Yokokawa (RIKEN).