Use this search facility to find out more about the profile of our HPC-Europa2 visitors, the type of work they have been doing, and their project achievements.

Abstract
Objectives
We aimed to carry out ab-initio calculations (in the framework of the Density Functional Theory) on silicon-germanium nanowires (SiGe NWs) doped with boron (B) and phosphorous (P) impurities and oriented along the [110] direction. The main motivation of the project was related to the crucial and decisive role that doping has in all the technological and device applications of SiGe NWs [1,2] and to the paucity of theoretical efforts on doped SiGe NWs’ s structural, electronic and transport properties. The main objectives of the project were the following: i) A complete comprehension of the role of B and P impurities on the structural and geometrical stability of the SiGe NWs, with particular emphasis on the different structural changes that B and P atoms can induce in different region of the wire (i.e. Si or Ge region). ii) To obtain a complete description about the thermodynamical stability of the single-doped and co-doped structures in order to give useful informations for the synthesis of these type of wires. This objective can be reached by calculating the Formation Enthalpy of the NWs and evaluating how this quantity is affected by the size’s wire, by the position of the impurities in the wire, by the relative distance impurity-impurity in the unit cell. iii) To obtain a complete description of role of doping and co-doping on the electronic properties of the wires; this can be reached by calculating band structure and density of states of the single-doped and co-doped NWs and comparing them with the undoped ones. iv) To analyze the nature of spatial wave function localization in order to show the possible separation between the impurity and the donor carrier. v) To calculate, through the Landauer approach, the transport properties of the NWs, pointing out how much crucial is the role of the impurity on the electronic mobility. We intended to perform all the calculations with the SIESTA code for structural and electronic calculations and with the TRANSIESTA code for the transport calculations, in particular the optimized version especially reengineered for the massively parallel machines of the Barcelona Supercomputing Centre (BSC), which could have permitted us to satisfy our computational requirements. To perform simulations which can be more close to the reality of the experiment our purpose was to study NWs with diameters ranging from 1.6 nm to 3 nm, that means systems with a unit cell containing nearly two hundreds atoms; we knew that this aspect would have required a very huge computational demand, that only with a high performance computing facility could have been satisfied. Moreover in order to analyze the effect of doping for these type of systems, we have intended to perform calculations with a unit cell that is four, five or six times the unit cell of the undoped case; this choice was made in order to minimize all the possible interactions between the impurities and to try to make the concentration of the impurity more close to the experiment; moreover, for each SiGe NWs, our aim was to study the effect of B-doping, P-doping and co-doping in the Si region and in the Ge region of the wire. This aspect should have increased the number of configurations on which performing calculations, making this part of the work the most demanding from a computational point of view; infact the considered unit cells were made of more than six hundreds atoms, which means that we should have used a very large number of processors in order to reduce as much as possible the CPU time. We applied for a six weeks HPC-Europa2 visit, which has been split in two shorter ones, in order to satisfy the demands of the two groups involved.

References: [1] J. Xiang et al., Nature 441, 489 (2005); [2] L. Lauhon et al., Nature 420, 57 (2002)

Achievements
We can state that the main objectives of the project have been successfully reached. We have performed first-principles calculations (based on Density Functional Theory and using the SIESTA code) concerning SiGe nanowires (SiGe NWs), doped with B and P impurities, oriented along the [110] direction and with diameters of 1.6 and 2.4 nm. We have focused on the core-shell geometry which recently has been the most interesting for all the technological and device applications of SiGe NWs. In particular we have analyzed how the effect of an impurity can modify the structural and electronic properties of a core-shell SiGe NW. The analysis of the structural properties of a B and P doped SiGe NW has been carried out analyzing the energetics of an impurity substitution in the core and in the shell of Sicore/Geshell (Si/Ge) and Gecore/Sishell (Ge/Si) NWs and evaluating the formation energy for each case. Since the the definition of formation energy (FE) of an impurity for a coaxial nanowire is not straightforward and there is a paucity of theoretical efforts on this argument, we have deeply studied the phenomena developing a new theoretical formulation for FE of doped core-shell SiGe NWs, starting from the generalization of Zhang and Northrup formula for reduced dimensionality system. Our results show that given a certain coaxial structure (Ge/Si or Si/Ge) the impurities prefer to occupy lattice sites located into the wire shell, pointing out the central role of geometry for this type of quantity. The only exception to this behavior is the P-doping of Si/Ge, where the Si-core doping is favored, therefore favoring a particular type of chemical bonding (Si-P bonds) with respect to the geometry of the system. Maybe the origin of this result can be ascribed to the fact that for the FE of a P impurity into a pure Si NW is smaller than the one in a pure Ge NW. As regards as the electronic properties we have analyzed B and P doping into the core and into the shell for both the types of core-shell NWs. In particular we have analyzed how the impurity type (B or P) and position (core or shell) can modify band structure and wave function localization given a certain coaxial structure. The results of our calculations show how, in contrast with B and P doping of pure Si and pure Ge NWs (for which several studies have demonstrated that the doping mechanism is inefficient at very small diameters), an efficient n-type and p-type doping can be reached in the case of core-shell SiGe NWs; this very interesting result is a consequence of the type II band offset between Si and Ge, which implies that valence states are on the germanium part of the wire while conduction states are on the silicon one. In particular in the case of Ge/Si NW with a P impurity into the core the impurity level falls inside the conduction band, yielding an electron at its bottom. This result is related to the type II band-offset that comes out at Si/Ge interface. In a pure Ge NW of the same size of the core, the impurity level would have been deep into the band gap and very difficult to activate at typical device temperatures. In this case instead, analyzing the wave function localization, we have found that the bottom of conduction band is on silicon-shell, which are below the impurity and all the germanium-core states; it means that the impurity does not need to be thermally activated. This indicates the formation of a one-dimensional electron gas that can have relevant importance for device applications. Other interesting case is the configuration with a B atom into the silicon shell of a Ge/Si NW. Again the impurity level is deep into the valence states and we have the formation of one-dimensional hole gas. These types of results can be easily extended to the case of Si/Ge NWs. The demonstration of an efficient doping for these types of wires can open substantial opportunities for the understanding doping mechanism at nanoscale and for improving its technological application.

Abstract
Objectives
The Traveling Salesman problem (TSP) is a combinatorial optimization problem that belongs to the NP-complete class. First formulated in 1930, is one of the most intensively studied benchmarks in optimization. Given a collection of cities and the cost of travel between each pair of them, the TSP is to find the cheapest way of visiting all of the cities and returning to the starting point. In the standard version, the travel costs are symmetric in the sense that traveling from city x to city y costs just as much as traveling from y to x. It is known as Symmetric TSP. In February 2009, Robert Bosch created a 100,000-city instance of the Symmetric Traveling Salesman Problem (TSP) that provides a representation of Leonardo da Vinci's Mona Lisa as a continuous-line drawing. Techniques for developing such point sets have evolved over the past several years through work of Robert Bosch and Craig Kaplan. An optimal solution to the 100,000-city Mona Lisa instance would set a new world record for the TSP. We have studied the solution of Monalisa TSP Challenge with Genetic Algoritmhs and Ant Colony Optimization. Employed methods and results are discussed in this work.

Achievements
At the moment we are obtaining solutions to Monalisa TSP problems that are close to the theoretical optimum, at a rate of 90%. We expect to improve this results during the next months.

Abstract
Objectives
Most scanners for small animal are based on cone-beam geometry with flat detector orbiting on a circular path. The reconstruction of these systems is usually done with a method based on the algorithm proposed by Feldkamp, Davis and Kress (FDK). The speed increase in the reconstruction for X-ray tomography (CT) is a prerequisite for the expansion of its clinical application. This paper presents an efficient implementation of a reconstruction algorithm FDK-based, modular, which uses the possibilities of parallel computing and efficient interpolation provided in CUDA using texture memory offered by the graphics-processing unit (GPU). The algorithm, tested on a micro-high-resolution CT, has improved the execution speed of a rear-projection stage 40x factor over a sequential implementation of reference written in C, remaining at all times the quality of the reconstruction.

Achievements
This paper presents a solution to the problem of accelerated reconstruction for FDK. It provides a modular implementation of the stages of filtering and back projection, which improves time on a 25x and 162X respectively. Although it is difficult to make direct comparisons between different implementations due mainly to differences between the hardware, we can say that the proposed implementation is an improvement over recently published a 4x. The main disadvantage of making filtered back projection with different modules is the inability to exploit synergies between the different stages, leading to a significant increase in transfers between CPU and GPU. However, the modularity allows efficient replacement algorithms implemented, facilitating the adaptability of the proposed solution to new architectures and devices.

The chosen strategy is to store the entire volume loaded in main memory and on the projections. This technique significantly reduces processing time and increase the locality of data stored in memory.

The need to interpolate the values of the projection makes the projections are ideal candidates to be stored in texture memory. In this way you get two benefits. On the one hand, the cost of collection is roughly equivalent to a simple reading, saving, the calculation of the interpolation. On the other hand, the cache memory access speeds texture continued access to data physically close.