Performance Evaluation of 3D Optoelectronic Computer Architectures Based on the FFT and Sorting Benchmarks

摘要

Optical interconnections can achieve greater data rates and lower power consumption than electronic. Yet, no computers featuring such interconnections have been demonstrated. It is projected that the first such computers will be composed of a stack of arrays of electronic processing elements, each one connected to its neighbors using free space optics. This architecture is called a 3D optoelectronic computer. Due to the SIMD nature of the processing element arrays, this architecture is be better suited for problems requiring massive parallelism. In this paper, we use a set of FFT and integer sorting benchmarks to evaluate the performance of such architectures and compare it with the best parallel supercomputers. For each benchmark we develop a suitable architecture and implementation of the algorithm. We calculate its performance based on the communication data rate between neighboring arrays and the performance of the electronic processing elements. We show that such architectures can outperform electronic supercomputers in both cases.