Performance Evaluations of Finite Difference Applications Realized on a Single Flux Quantum Circuits-Based Reconfigurable Accelerator

Abstract—Hardware accelerators integrating to general purpose processors are increasingly employed to achieve lower power consumption and higher processing speed, however, energy consumption of high performance accelerators has become a great issue on large scale parallel computer system. We have investigated the applicability of Single-Flux-Quantum (SFQ) circuits as a part of superconductivity technology in high-performance computing systems. Although it is possible to develop extraordinary low power processor by SFQ devices, conditional branch and loop back controls are difficult to be implemented by current SFQ technology. Therefore, we have proposed Reconfigurable DataPath (RDP) accelerator which is avoiding those limitations of SFQ technology, while trying to get benefits of these circuits. In this research, we have implemented two-dimensional Heat (2D-Heat) and Finite Difference Time Domain (2D-FDTD) applications for investigating efficiency of using SFQ-RDP accelerator. According to performance evaluation results for above applications, execution times are 50.6 and 79.0 times smaller than those of the general purpose processor, and comparable with ones reported for GPU (Graphics Processing Units).Hardware accelerators integrating to general purpose processors are increasingly employed to achieve lower power consumption and higher processing speed, however, energy consumption of high performance accelerators has become a great issue on large scale parallel computer system. We have investigated the applicability of Single-Flux-Quantum (SFQ) circuits as a part of superconductivity technology in high-performance computing systems. Although it is possible to develop extraordinary low power processor by SFQ devices, conditional branch and loop back controls are difficult to be implemented by current SFQ technology. Therefore, we have proposed Reconfigurable Data-Path (RDP) accelerator which is avoiding those limitations of SFQ technology, while trying to get benefits of these circuits. In this research, we have implemented two-dimensional Heat (2D-Heat) and Finite Difference Time Domain (2D-FDTD) applications for investigating efficiency of using SFQ-RDP accelerator. According to performance evaluation results for above applications, execution times are 50.6 and 79.0 times smaller than those of the general purpose processor, and comparable with ones reported for GPU (Graphics Processing Units).