ESC-SV Special: Multicore and Massively Parallel Platforms and Moore's Law Scalability

A number of parallel processing platforms are entering mainstream use in embedded systems. SMP (symmetric multiprocessing) multithreaded architectures, now common in desktops and servers, and SIMD (single-instruction, multiple data) architectures from the supercomputing and graphics worlds, are well-known. New massively parallel processor array (MPPA) architectures, specifically aimed at high-performance media, imaging, networking and wireless applications, have recently appeared. This paper considers parallel platform candidates for embedded systems, in the context of pre-existing single-processor, DSP, FPGA and ASIC technologies.

Each platform is presented, and its characteristics, benefits and costs are discussed. Particular emphasis is given to the long-term scalability of all these platform choices as Moore's Law continues to moves forward.

Multicore today means a small number of processors, from two to eight. This paper uses massively parallel to mean hundreds of processors or more.