Sparse matrix multiplication on a reconfigurable many-core architecture

Sparse matrix-vector multiplication (SMVM) is a fundamental operation in many scientific and engineering applications. In many cases sparse matrices have thousands of rows and columns where most of the entries are zero, while non-zero data is spread over the matrix. This sparsity of data locality reduces the effectiveness of data cache in general-purpose processors quite reducing their performance efficiency when compared to what is achieved with dense matrix multiplication. In this paper, we propose a parallel processing solution for SMVM in a many-core architecture. The architecture is tested with known benchmarks using a ZYNQ-7020 FPGA. The architecture is scalable in the number of core elements and limited only by the available memory bandwidth. It achieves performance efficiencies up to almost 70% and better performances than previous FPGA designs.

Algorithm-oriented design of efficient many-core architectures applied to dense matrix multiplication

Recent integrated circuit technologies have opened the possibility to design parallel architectures with hundreds of cores on a single chip. The design space of these parallel architectures is huge with many architectural options. Exploring the design space gets even more difficult if, beyond performance and area, we also consider extra metrics like performance and area efficiency, where the designer tries to design the architecture with the best performance per chip area and the best sustainable performance. In this paper we present an algorithm-oriented approach to design a many-core architecture. Instead of doing the design space exploration of the many core architecture based on the experimental execution results of a particular benchmark of algorithms, our approach is to make a formal analysis of the algorithms considering the main architectural aspects and to determine how each particular architectural aspect is related to the performance of the architecture when running an algorithm or set of algorithms. The architectural aspects considered include the number of cores, the local memory available in each core, the communication bandwidth between the many-core architecture and the external memory and the memory hierarchy. To exemplify the approach we did a theoretical analysis of a dense matrix multiplication algorithm and determined an equation that relates the number of execution cycles with the architectural parameters. Based on this equation a many-core architecture has been designed. The results obtained indicate that a 100 mm(2) integrated circuit design of the proposed architecture, using a 65 nm technology, is able to achieve 464 GFLOPs (double precision floating-point) for a memory bandwidth of 16 GB/s. This corresponds to a performance efficiency of 71 %. Considering a 45 nm technology, a 100 mm(2) chip attains 833 GFLOPs which corresponds to 84 % of peak performance These figures are better than those obtained by previous many-core architectures, except for the area efficiency which is limited by the lower memory bandwidth considered. The results achieved are also better than those of previous state-of-the-art many-cores architectures designed specifically to achieve high performance for matrix multiplication.

Predictive Optimal Matrix Converter Control for a Dynamic Voltage Restorer with Flywheel Energy Storage

This paper presents a predictive optimal matrix converter controller for a flywheel energy storage system used as Dynamic Voltage Restorer (DVR). The flywheel energy storage device is based on a steel seamless tube mounted as a vertical axis flywheel to store kinetic energy. The motor/generator is a Permanent Magnet Synchronous Machine driven by the AC-AC Matrix Converter. The matrix control method uses a discrete-time model of the converter system to predict the expected values of the input and output currents for all the 27 possible vectors generated by the matrix converter. An optimal controller minimizes control errors using a weighted cost functional. The flywheel and control process was tested as a DVR to mitigate voltage sags and swells. Simulation results show that the DVR is able to compensate the critical load voltage without delays, voltage undershoots or overshoots, overcoming the input/output coupling of matrix converters.

Direct Power Control of Matrix Converter Based Unified Power Flow Controllers

This paper presents the Direct Power Control of Three-Phase Matrix Converters (DPC-MC) operating as Unified Power Flow Controllers (UPFC). Since matrix converters allow direct AC/AC power conversion without intermediate energy storage link, the resulting UPFC has reduced volume and cost, together with higher reliability. Theoretical principles of DPC-MC method are established based on an UPFC model, together with a new direct power control approach based on sliding mode control techniques. As a result, active and reactive power can be directly controlled by selection of an appropriate switching state of matrix converter. This new direct power control approach associated to matrix converters technology guarantees decoupled active and reactive power control, zero error tracking, fast response times and timely control actions. Simulation results show good performance of the proposed system.

Grid structure impact in sparse point representation of derivatives

In the Sparse Point Representation (SPR) method the principle is to retain the function data indicated by significant interpolatory wavelet coefficients, which are defined as interpolation errors by means of an interpolating subdivision scheme. Typically, a SPR grid is coarse in smooth regions, and refined close to irregularities. Furthermore, the computation of partial derivatives of a function from the information of its SPR content is performed in two steps. The first one is a refinement procedure to extend the SPR by the inclusion of new interpolated point values in a security zone. Then, for points in the refined grid, such derivatives are approximated by uniform finite differences, using a step size proportional to each point local scale. If required neighboring stencils are not present in the grid, the corresponding missing point values are approximated from coarser scales using the interpolating subdivision scheme. Using the cubic interpolation subdivision scheme, we demonstrate that such adaptive finite differences can be formulated in terms of a collocation scheme based on the wavelet expansion associated to the SPR. For this purpose, we prove some results concerning the local behavior of such wavelet reconstruction operators, which stand for SPR grids having appropriate structures. This statement implies that the adaptive finite difference scheme and the one using the step size of the finest level produce the same result at SPR grid points. Consequently, in addition to the refinement strategy, our analysis indicates that some care must be taken concerning the grid structure, in order to keep the truncation error under a certain accuracy limit. Illustrating results are presented for 2D Maxwell's equation numerical solutions.

Power converter topologies for wind energy conversion systems: Integrated modeling, control strategy and performance simulation

This paper presents new integrated model for variable-speed wind energy conversion systems, considering a more accurate dynamic of the wind turbine, rotor, generator, power converter and filter. Pulse width modulation by space vector modulation associated with sliding mode is used for controlling the power converters. Also, power factor control is introduced at the output of the power converters. Comprehensive performance simulation studies are carried out with matrix, two-level and multilevel power converter topologies in order to adequately assert the system performance. Conclusions are duly drawn.

Impact of flavor-changing neutral current top quark interactions on BR(t -> bW)

We study the effect that flavor-changing neutral current interactions of the top quark will have on the branching ratio of charged decays of the top quark. We have performed an integrated analysis using Tevatron and B-factories data and with just the further assumption that the Cabibbo-Kobayashi-Maskawa matrix is unitary, we can obtain very restrictive bounds on the strong and electroweak flavor-changing neutral current branching ratios Br(t -> qX)< 4.0x10(-4), where X is any vector boson and a sum in q=u, c is implied.

Matrix Converter-Based Unified Power-Flow Controllers: Advanced Direct Power Control Method

This paper presents a direct power control (DPC) for three-phase matrix converters operating as unified power flow controllers (UPFCs). Matrix converters (MCs) allow the direct ac/ac power conversion without dc energy storage links; therefore, the MC-based UPFC (MC-UPFC) has reduced volume and cost, reduced capacitor power losses, together with higher reliability. Theoretical principles of direct power control (DPC) based on sliding mode control techniques are established for an MC-UPFC dynamic model including the input filter. As a result, line active and reactive power, together with ac supply reactive power, can be directly controlled by selecting an appropriate matrix converter switching state guaranteeing good steady-state and dynamic responses. Experimental results of DPC controllers for MC-UPFC show decoupled active and reactive power control, zero steady-state tracking error, and fast response times. Compared to an MC-UPFC using active and reactive power linear controllers based on a modified Venturini high-frequency PWM modulator, the experimental results of the advanced DPC-MC guarantee faster responses without overshoot and no steady-state error, presenting no cross-coupling in dynamic and steady-state responses.

Simulation of the Magnetic Induction Vector of a Magnetic Core to be Used in FFC NMR Relaxometry

This paper is a contribution for the assessment and comparison of magnet properties based on magnetic field characteristics particularly concerning the magnetic induction uniformity in the air gaps. For this aim, a solver was developed and implemented to determine the magnetic field of a magnetic core to be used in Fast Field Cycling (FFC) Nuclear Magnetic Resonance (NMR) relaxometry. The electromagnetic field computation is based on a 2D finite-element method (FEM) using both the scalar and the vector potential formulation. Results for the magnetic field lines and the magnetic induction vector in the air gap are presented. The target magnetic induction is 0.2 T, which is a typical requirement of the FFC NMR technique, which can be achieved with a magnetic core based on permanent magnets or coils. In addition, this application requires high magnetic induction uniformity. To achieve this goal, a solution including superconducting pieces is analyzed. Results are compared with a different FEM program.

Magnetic and plagioclase linear fabric discrepancy in dykes: a new way to define the flow vector using magnetic foliation

In basaltic dykes the magnetic lineation K1 (maximum magnetic susceptibility axis) is generally taken to indicate the flow direction during solidification of the magma. This assumption was tested in Tertiary basaltic dykes from Greenland displaying independent evidence of subhorizontal flow. The digital processing of microphotographs from thin sections cut in (K1, K2) planes yields the preferred linear orientation of plagioclase, which apparently marks the magma flow lineation. In up to 60% of cases, the angular separation between K1 and the assumed flow direction is greater than 45degrees. This suggests that the uncorroborated use of magnetic lineations in dykes is risky. A simple geometrical method is proposed to infer the flow vector from AMS in dykes based solely on magnetic foliations.

Removing ambiguities in the neutrino mass matrix

We suggest that the weak-basis independent condition det(M-nu) = 0 for the effective neutrino mass matrix can be used in order to remove the ambiguities in the reconstruction of the neutrino mass matrix from input data available from present and future feasible experiments. In this framework, we study the full reconstruction of M-nu with special emphasis on the correlation between the Majorana CP-violating phase and the various mixing angles. The impact of the recent KamLAND results on the effective neutrino mass parameter is also briefly discussed. (C) 2003 Elsevier Science B.V. All rights reserved.

Inverse dynamics based predictive control for unified power flow controllers without DC bus

This paper presents a new predictive digital control method applied to Matrix Converters (MC) operating as Unified Power Flow Controllers (UPFC). This control method, based on the inverse dynamics model equations of the MC operating as UPFC, just needs to compute the optimal control vector once in each control cycle, in contrast to direct dynamics predictive methods that needs 27 vector calculations. The theoretical principles of the inverse dynamics power flow predictive control of the MC based UPFC with input filter are established. The proposed inverse dynamics predictive power control method is tested using Matlab/Simulink Power Systems toolbox and the obtained results show that the designed power controllers guarantees decoupled active and reactive power control, zero error tracking, fast response times and an overall good dynamic and steady-state response.

Linear and sliding-mode control design for matrix converter-based unified power flow controllers

This paper presents the design and compares the performance of linear, decoupled and direct power controllers (DPC) for three-phase matrix converters operating as unified power flow controllers (UPFC). A simplified steady-state model of the matrix converter-based UPFC fitted with a modified Venturini high-frequency pulse width modulator is first used to design the linear controllers for the transmission line active (P) and reactive (Q) powers. In order to minimize the resulting cross coupling between P and Q power controllers, decoupled linear controllers (DLC) are synthesized using inverse dynamics linearization. DPC are then developed using sliding-mode control techniques, in order to guarantee both robustness and decoupled control. The designed P and Q power controllers are compared using simulations and experimental results. Linear controllers show acceptable steady-state behaviour but still exhibit coupling between P and Q powers in transient operation. DLC are free from cross coupling but are parameter sensitive. Results obtained by DPC show decoupled power control with zero error tracking and faster responses with no overshoot and no steady-state error. All the designed controllers were implemented using the same digital signal processing hardware.

The extended unsymmetric frontal solution for multiple-point constraints

The purpose of this paper is to discuss the linear solution of equality constrained problems by using the Frontal solution method without explicit assembling. Design/methodology/approach - Re-written frontal solution method with a priori pivot and front sequence. OpenMP parallelization, nearly linear (in elimination and substitution) up to 40 threads. Constraints enforced at the local assembling stage. Findings - When compared with both standard sparse solvers and classical frontal implementations, memory requirements and code size are significantly reduced. Research limitations/implications - Large, non-linear problems with constraints typically make use of the Newton method with Lagrange multipliers. In the context of the solution of problems with large number of constraints, the matrix transformation methods (MTM) are often more cost-effective. The paper presents a complete solution, with topological ordering, for this problem. Practical implications - A complete software package in Fortran 2003 is described. Examples of clique-based problems are shown with large systems solved in core. Social implications - More realistic non-linear problems can be solved with this Frontal code at the core of the Newton method. Originality/value - Use of topological ordering of constraints. A-priori pivot and front sequences. No need for symbolic assembling. Constraints treated at the core of the Frontal solver. Use of OpenMP in the main Frontal loop, now quantified. Availability of Software.

The norm of the K- Th derivative of the X-symmetric power of an operator

In this paper, the exact value for the norm of directional derivatives, of all orders, for symmetric tensor powers of operators on finite dimensional vector spaces is presented. Using this result, an upper bound for the norm of all directional derivatives of immanants is obtained.