Design Philosophy of a Networking-Oriented Data-Driven Processor: CUE

Hiroaki NISHIKAWA  

Publication
IEICE TRANSACTIONS on Electronics   Vol.E89-C   No.3   pp.221-229
Publication Date: 2006/03/01
Online ISSN: 1745-1353
Print ISSN: 0916-8516
Type of Manuscript: INVITED PAPER (Special Section on VLSI Design Technology in the Sub-100 nm Era)
Category: 
Keyword: 
data-driven,  multiprocessing,  real-time,  VLSI,  

Full Text: PDF(1021.9KB)
>>Buy this Article


Summary: 
To realize a secure networking infrastructure, the author is carrying out CUE (Coordinating Users' requirements and Engineering constraints) project with a network carrier and a VLSI manufacture. Since CUE-series data-driven processors developed in the project were specifically designed to be an embedded programmable component as well as a multi-processor element, particular design considerations were taken to achieve real-time multiprocessing capabilities essentially needed in multi-media communication environment. A novel data-driven paradigm is first introduced with special emphasis on VLSI-oriented parallel processing architectures. Data-driven protocol handlings on CUE-p and CUE-v1 are then discussed for their real-time multiprocessing capability without any runtime overheads. The emulation facility RESCUE (Real-time Execution System for CUE-series data-driven processors) was also built to develop scalable chip multi-processors in self-evolutional manner. Based on emulation results, the latest version named CUE-v2 was realized as a hybrid processor enabling simultaneous processing of data-driven and control-driven threads to achieve higher performance for inline processing and to avoid any bottlenecks in sequential parts of real-time programs frequently encountered in actual time-sensitive applications. Effectiveness of the data-driven chip multi-processor architecture will finally be addressed for lower power consumption and scalability to realize future VLSI processors in the sub-100 nm era.