On kernel acceleration of electromagnetic solvers via hardware emulation. (October 2015)

• Record Type: Journal Article
• Authors: Tarek Ibn Ziad, M.; Hossam, Mohamed; Masoud, Mohamad A.; Nagy, Mohamed; Adel, Hesham A.; Alkabani, Yousra; El-Kharashi, M. Watheq; Salah, Khaled; AbdelSalam, Mohamed
• Is Part Of: Computers & electrical engineering. Volume 47(2015)
• Page Start: 96
• Page End: 113
• Publication Date: 2015-10
• Subjects: Electromagnetic simulations -- Emulation in hardware -- Finite element method -- Gaussian Elimination method -- Jacobi iterative method -- Metamaterials; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2015.08.017
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: We present an emulation-based accelerator for EM simulations in metamaterials. We propose two hardware designs to solve systems of linear equations from the FEM. Both presented designs are run on a physical Veloce 1 emulator from Mentor Graphics. Timing results exceed the performance of several software-based solvers. Graphical abstract: Abstract: Finding new techniques to accelerate electromagnetic (EM) simulations has become a necessity nowadays due to its frequent usage in industry. As they are mainly based on domain discretization, EM simulations require solving enormous systems of linear equations simultaneously. Available software-based solutions do not scale well with the increasing number of equations to be solved. As a result, hardware accelerators have been utilized to speed up the process. We introduce using hardware emulation as an efficient solution for EM simulation core solvers. Two different scalable architectures are implemented to accelerate the solver part of an EM simulator based on the Gaussian Elimination and the Jacobi iterative methods. Results show that the performance gap between presented solutions and software-based ones increases as the number of equations increases. For example, solving 2, 002, 000 equations using our Clustered Jacobi design in single floating-point precision achieved a speed-up of 100.88x and 35.24x over pure software implementations represented by MATLAB and the ALGLIB C++ package, respectively.