Xilinx, best known as a supplier of field-programmable gate arrays (FPGAs), is increasingly a supplier of SoCs that are predominantly software programmable but with some FPGA fabric and programmable I/O under its Zync brand.
The Zynq MPSOC is implemented in 16-nm 16FF+ manufacturing process from TSMC and began shipping Sept. 30, 2015. The chip includes four Cortex-A53 cores and two Cortex-R5 cores together with Mali GPU, security cores and accelerators and an HEVC codec that supports graphics transmission and reception up to 4K image resolution.
The level of performance and graphics capability make the chip family suitable for demanding applications in machine vision such as autonomous driver assistance systems (ADAS), said Steve Glaser, senior vice president of corporate strategy and marketing. In addition, applications that require security support or data fusion are suited to the chip family. An expanded range of software library offerings are available in support of neuromorphic computing from Auviz, MulticoreWare and TeraDeep.
The ecosystem includes hardware design IP providers in the areas of industrial Ethernet, video and connectivity, codecs, and there are also numerous design services providers to help provide users with both performance and differentiation through programmability. Speaking at the Embedded World exhibition Glaser told eeNews Europe that machine learning algorithms implemented in FPGA fabric on the Zynq MPSOC are running 50 times faster than the same algorithms running on the CPUs and 5 to 10 times faster than the algorithms running on GPUs.
Glaser added that SDX, Xilinx’s software-defined development environments, would be evolving into more application-specific versions in the future, such as embedded vision development.